Post on 23-Apr-2022
DAMAGE CHARACTERIZATION OF FIBER REINFORCED COMPOSITE
MATERIALS BY MEANS OF MULTIAXIAL TESTING
AND DIGITAL IMAGE CORRELATION
by
Joseph Terrance Jette
A thesis submitted in partial fulfillment
of the requirements for the degree
of
Master of Science
in
Mechanical Engineering
MONTANA STATE UNIVERSITY
Bozeman, Montana
August 2017
©COPYRIGHT
by
Joseph Terrance Jette
2017
All Rights Reserved
ii
ACKNOWLEDGEMENTS
There have been many people throughout this process that helped me a great deal.
First off, I would like to thank my committee; Dr. Douglas Cairns, Dr. David Miller, and
Dr. Michael Edens have been indispensable to my academic success. I would especially
like to thank Dr. Douglas Cairns for encouraging me to pursue graduate school, Dr. David
Miller for his guidance, and Dr. Michael Edens for countless hours of work in the lab. I
would also like to thank Dan Samborsky for his assistance, guidance, and humor that
helped me tremendously. I would also like to thank Boeing for their contributions and
support of this research.
Many fellow graduate and undergraduate students have also been a tremendous
help to me. I would like to highlight Matt Peterson and Tiok Agastra’s contributions with
finite element modeling in ANSYS. I would also like to acknowledge Mike Voth, Ryan
Clarke, Cody Atwood, and Jake Nunemaker for their various contributions and aid.
Finally, I could have never gotten this far without the love and support from family
and friends. I would like to thank my mom, my dad, my two sisters, and my lovely wife
for their encouragement and support along the way.
iii
TABLE OF CONTENTS
1. INTRODUCTION ......................................................................................................13
Introduction to Composite Materials ......................................................................... 13 Brief History of Composite Materials ................................................................. 14 Composite Material Characterization .................................................................. 16 Damage and Failure in Composite Materials ...................................................... 19
Motivation .................................................................................................................. 21
2. THE MONTANA STATE UNIVERISTY IN-PLANE LOADER ..............................22
Multiaxial Testing ...................................................................................................... 22
Failure and Strength of Composites ..................................................................... 24 Maximum Stress Criterion ..............................................................................25 Maximum Strain Criterion ..............................................................................25
Interacting Criterion ........................................................................................26 Brief History of the IPL ............................................................................................. 29
Third Generation .................................................................................................. 30 Fourth Generation ................................................................................................ 32 Fifth Generation ................................................................................................... 32
Application of the IPL ............................................................................................... 38
Dissipated Energy Density ................................................................................... 38 Digital Image Correlation .................................................................................... 43
Test Coupon ............................................................................................................... 48
Material Systems .................................................................................................. 48 Geometry ............................................................................................................. 50
Manufacturing ...................................................................................................... 51
3. EXPERIMENTAL DATA – MULTIAXIAL TESTS ..................................................55
Testing Procedure ....................................................................................................... 55 Loading Paths ...................................................................................................... 57 Data Acquisition .................................................................................................. 61
Post-Processing in MATLAB .................................................................................... 64
Basic Post-Processing Scheme ............................................................................ 64 Results and Discussion .............................................................................................. 71
Failed Coupons .................................................................................................... 71
Digital Image Correlation Results ....................................................................... 75 Failure Surfaces ................................................................................................... 81
4. FINITE ELEMENT MODEL .....................................................................................94
Model Definition ........................................................................................................ 94 Model and Layup Setup ....................................................................................... 95
iv
TABLE OF CONTENTS – CONTINUED
Module A. Geometry ......................................................................................95 Module B. Engineering Data ..........................................................................96 Module C. ACP (Prep) ....................................................................................97 Module D. Static Structural ............................................................................98
Module E. ACP (Post) .....................................................................................99 Parameter Set ................................................................................................100
Boundary Conditions ......................................................................................... 100 Model Assumptions ........................................................................................... 101
Processing ................................................................................................................ 103 Results and Discussion ............................................................................................ 105
5. CONCLUSIONS ....................................................................................................... 112
Experimental Conclusions ........................................................................................ 113
Finite Element Model Conclusions ........................................................................... 116
6. FUTURE WORK ...................................................................................................... 119
High Strain-Rate Multiaxial Testing ......................................................................... 119
Multiaxial Testing Recommendations ...................................................................... 119
Data Processing ................................................................................................... 119 IPL Compliance ................................................................................................. 121 Automation of the IPL ....................................................................................... 122
IPL Software and Control .................................................................................. 122 Long-Term Suggestions ..................................................................................... 124
Digital Image Correlation (ARAMIS) Recommendations ...................................... 125 Computer Resources .......................................................................................... 125
Software Limitations .......................................................................................... 126 Camera Hardware .............................................................................................. 126
REFERENCES CITED ....................................................................................................128
APPENDICES .................................................................................................................132
APPENDIX A: Test Matrices ............................................................................. 133 APPENDIX B: Laminate-Level Failure ............................................................. 140 APPENDIX C: Progressive Damage Tables ....................................................... 145
v
LIST OF TABLES
Table Page
1. Lamina material properties for both material systems. Material
properties are specified for material at room temperature and
dry (RTD) [18] ................................................................................................ 48
2. Strength properties for IM7/8552. Properties are specified for
material under quasistatic loading at room temperature and dry
(RTD) [18] ...................................................................................................... 49
3. Laminate stacking sequence table and layup numbers. Naming
convention of tests uses "Layup Number" to replace XX of
XX_0YY ......................................................................................................... 50
4. Statistical values for geometry of all tested coupons. Measured
with Mitutoyo digital calipers verified with standardized
machinist gauge blocks ................................................................................... 53
5. Snippet of progressive damage table as developed from the
images above. For the full table and for each laminate, refer to
APPENDIX C ................................................................................................. 79
6. Test Matrix for MSU-11. [-45/90/45/0]s laminate with
IM7/8552 material system. Comp. abbreviation for
compression .................................................................................................. 134
7. Test Matrix for MSU-13. [0/90/0/90]s laminate with IM7/8552
material system. Comp. abbreviation for compression. NFF
abbreviation for No Final Failure. GS abbreviation for Grip
Slippage......................................................................................................... 135
8. Test Matrix for MSU-14. [-45/45/-45/45]s laminate with
IM7/8552 material system. Comp. abbreviation for
compression. NFF abbreviation for No Final Failure. GS
abbreviation for Grip Slippage. IPL Lim. Abbreviation for IPL
Limit Reached ............................................................................................... 136
9. Test Matrix for MSU-1. [-45/90/45/0]s laminate with Toray
material system. Comp. abbreviation for compression. NFF
abbreviation for No Final Failure. GS abbreviation for Grip
Slippage......................................................................................................... 137
vi
LIST OF TABLES – CONTINUED
Table Page
10. Test Matrix for MSU-3. [0/90/0/90]s laminate with Toray
material system. Comp. abbreviation for compression. NFF
abbreviation for No Final Failure. GS abbreviation for Grip
Slippage. IPL Lim. Abbreviation for IPL Limit Reached ............................. 138
11. Test Matrix for MSU-3. [-45/45/-45/45]s laminate with Toray
material system. Comp. abbreviation for compression. NFF
abbreviation for No Final Failure. GS abbreviation for Grip
Slippage. IPL Lim. Abbreviation for IPL Limit Reached ............................. 139
12. Progressive damage table for MSU–11 tests. IM7/8552
Material: [-45/90/45/0]s ................................................................................ 147
13. Progressive damage table for MSU–13 tests. IM7/8552
Material: [0/90/0/90]s ................................................................................... 148
14. Progressive damage table for MSU–14 tests. IM7/8552
Material: [-45/45/-45/45]s ............................................................................. 149
15. Progressive damage table for MSU–1 tests. Toray Material: [-
45/90/45/0]s .................................................................................................. 150
16. Progressive damage table for MSU–3 tests. Toray Material:
[0/90/0/90]s ................................................................................................... 151
17. Progressive damage table for MSU–4 tests. Toray Material: [-
45/45/-45/45]s ............................................................................................... 152
vii
LIST OF FIGURES
Figure Page
1. The "building-block approach" schematic as shown in MIL-
HDBK-17-1F [4] ............................................................................................. 18
2. Intra-ply damages in a composite material (by Anderson [6]).
(1) Fiber Pull-out. (2) Fiber Bridging. (3) Debonding. (4) Fiber
Breakage / Rupture. (5) Matrix Plasticity & Cracking ................................... 20
3. Schematic of possible deformations applied to a coupon in the
IPL................................................................................................................... 29
4. A photograph of the first-generation IPL as shown in Ritter's
thesis [12] ........................................................................................................ 30
5. Fifth-generation Montana State University In-Plane Loader. A:
Out-of-plane Constrainers (both sides) B: Coupon Loading
Location C: Main Electronics ......................................................................... 33
6. Solid model rendering of the fifth-generation IPL grip
assembly. A: Hydraulic Piston B: Carbide-textured Grip Plates
C: Transverse Support Plates .......................................................................... 35
7. Latest vector loop schematic and nomenclature as used for IPL
control software .............................................................................................. 37
8. A generic load vs displacement plot as seen in Collett [11].
Only to illustrate dissipated energy ................................................................. 38
9. Example dissipated energy calculation for sample 11_008.
Dissipated energy is expressed in units of 𝑙𝑏𝑓 ∙ 𝑖𝑛 ......................................... 42
10. Image of acceptable stochastic pattern with 15×15 facets [17] ..................... 45
11. Facet tracking as shown in the ARAMIS manual [17] ................................... 45
12. Example of ARAMIS post-processed data and data
presentation. (Still image of a video taken just after damage
initiation) ......................................................................................................... 46
13. GOM ARAMIS hardware as shown in ARAMIS Manual [17] ...................... 47
viii
LIST OF FIGURES – CONTINUED
Figure Page
14. Latest coupon geometry. Identical gauge section to previous
samples but larger grip areas. The displayed coordinate system
is not displayed at the working origin. Reference below, Figure
20..................................................................................................................... 51
15. Diamondlike-Coated Carbide End Mill, Ball-End, 4 Flute, 1/4"
Mill Diameter, 2-1/2" Overall Length [19] ..................................................... 52
16. Result of manufacturing procedure. *This sample was flawed
due to improper tabbing and was therefore not tested. A side-
view to show tabs (top). Front view (middle). Inch ruler for
scale (bottom).................................................................................................. 54
17. Schematic of NRL’s loading path definitions [9] ........................................... 57
18. Image of out-of-plane displacements recorded via DIC for a
failed compression test .................................................................................... 59
19. Normalized load paths (displacement paths) performed for
every laminate. Vectors shown as unit vectors to display
direction only .................................................................................................. 60
20. Left image of test 11_024 as a demonstration of ARAMIS
image acquisition. (Green line [MC]) shows "movement
correction". (Red line [VE 2]) shows “virtual extensometer”.
Coordinate system is shown at the notch tip ................................................... 63
21. Image of facet truncation. Area enclosed in the green rectangle
is the area of accepted facets ........................................................................... 66
22. Schematic view of the post-processing scheme used to further
process ARAMIS data via MATLAB. For more detail about
each step, refer to the section above ............................................................... 69
23. Laminate 11 interesting failures. Written labels correspond
directly to the test matrix. For scale, coupon widths are 1 inch
or refer to Figure 14 above .............................................................................. 72
24. Laminate 13 interesting failures. Written labels correspond
directly to the test matrix. For scale, coupon widths are 1 inch
or refer to Figure 14 above .............................................................................. 73
ix
LIST OF FIGURES – CONTINUED
Figure Page
25. Laminate 4 interesting failures. Written labels correspond
directly to the test matrix. For scale, coupon widths are 1 inch
or refer to Figure 14 above .............................................................................. 74
26. Test 11_021 image of Major Strain report at damage initiation ..................... 76
27. Test 11_021 image of Major Strain report at an arbitrary
intermediate damaged state ............................................................................. 77
28. Test 11_021 image of Major Strain report at the stage before
final failure ...................................................................................................... 78
29. Image depicting a selected facet from an arbitrary test as seen
in ARAMIS. "Stage Point Data" displays all facet properties ....................... 82
30. All in-plane strain data points collected for MSU - 11 tests.
This includes all tests displayed in the test matrices....................................... 83
31. MSU - 11 nested isosurfaces defined by data density. Scatter
points are also shown for reference ................................................................ 84
32. Example of nested isosurfaces displayed on the eps x = 0
primary plane .................................................................................................. 86
33. Nested isosurfaces for MSU - 11 displayed on each primary
plane. Original nested isosurfaces (top-left). Primary plane
eps xy = 0 (top-right). Primary plane eps y = 0 (bottom-left).
Primary plane eps x = 0 (bottom-right) .......................................................... 87
34. MSU - 1 nested isosurfaces defined by data density. Scatter
points are also shown for reference ................................................................ 88
35. Nested isosurfaces for MSU - 1 displayed on each primary
plane. Original nested isosurfaces (top-left). Primary plane
휀𝑥𝑦 = 0 (top-right). Primary plane 휀𝑦 = 0 (bottom-left).
Primary plane 휀𝑥 = 0 (bottom-right) ............................................................. 89
36. Direct comparison of MSU - 11 tests (Figure 33) shown in blue
and MSU - 1 tests (Figure 35) shown in red ................................................... 90
x
LIST OF FIGURES – CONTINUED
Figure Page
37. Ply-level "failure surfaces" produced from original point cloud
of the Hexcel IM7/8552 material system transformed and
resolved into each ply's local coordinate system. Original point
cloud (top-left). Primary plane 휀12 = 0 (top-right). Primary
plane 휀1 = 0 (bottom-left). Primary plane 휀2 = 0 (bottom-
right) ................................................................................................................ 92
38. Ply-level "failure surfaces" produced from original point cloud
of the Toray material system transformed and resolved into
each ply's local coordinate system .................................................................. 93
39. Top-level view of ANSYS Workbench model ................................................ 95
40. Model mesh definition with coordinate system shown ................................... 96
41. ACP (Prep) figure showing ply layer definition, coordinate
system, and distinction of elements through the model
thickness .......................................................................................................... 98
42. ACP (Post) example of failure display using Maximum Strain
criterion of an arbitrary simulation. Image only displays
predicted failure for selected layer (surface layer) ....................................... 100
43. Example of Displacements applied to top boundary condition
in ANSYS finite element model. Undeformed (top left),
Positive x-displacement (top right), Positive y-displacement
(bottom left), Positive rotation (bottom right) .............................................. 101
44. Example of the facets chosen for step 2 to determine boundary
conditions for the ANSYS model ................................................................. 104
45. An arbitrarily chosen sample (11_021) results compared
between experimental results and the finite element model.
Maximum principal strain displayed ............................................................ 106
46. The same sample shown in Figure 45 (11_021) results
compared between experimental results and the finite element
model. Shear elastic strain displayed ............................................................ 107
47. Point cloud developed from modeling all MSU - 11 tests and
extracting strain data. Displayed as before .................................................. 108
xi
LIST OF FIGURES – CONTINUED
Figure Page
48. Nested isosurfaces for FEM - 11 displayed on each primary
plane. Original nested isosurfaces (top-left). Primary plane
휀𝑥𝑦 = 0 (top-right). Primary plane 휀𝑦 = 0 (bottom-left).
Primary plane 휀𝑥 = 0 (bottom-right) ........................................................... 109
49. Direct comparison of "sliced" isosurfaces of MSU - 11 tests
(Figure 33Figure 35) in blue and the FEM of MSU - 11 (Figure
48) in red ........................................................................................................110
50. Third and Fourth generation IPL grip assembly as shown in
Collett [11]. LVDTs are used for the displacement control
software ......................................................................................................... 123
51. Nested isosurfaces for MSU - 13 displayed on each primary
plane. Original nested isosurfaces (top-left). Primary plane
eps xy = 0 (top-right). Primary plane eps y = 0 (bottom-left).
Primary plane eps x = 0 (bottom-right) ........................................................ 141
52. Nested isosurfaces for MSU - 14 displayed on each primary
plane. Original nested isosurfaces (top-left). Primary plane
eps xy = 0 (top-right). Primary plane eps y = 0 (bottom-left).
Primary plane eps x = 0 (bottom-right) ........................................................ 142
53. Nested isosurfaces for MSU - 3 displayed on each primary
plane. Original nested isosurfaces (top-left). Primary plane
eps xy = 0 (top-right). Primary plane eps y = 0 (bottom-left).
Primary plane eps x = 0 (bottom-right) ........................................................ 143
54. Nested isosurfaces for MSU - 4 displayed on each primary
plane. Original nested isosurfaces (top-left). Primary plane
eps xy = 0 (top-right). Primary plane eps y = 0 (bottom-left).
Primary plane eps x = 0 (bottom-right) ........................................................ 144
xii
ABSTRACT
Composite materials offer a unique quality to improve structural designs. Now, not
only can a structure’s geometry be designed, composite materials offer the engineer the
ability to design the layup of the material and, in turn, control some of its structural
properties. While this feature of composite materials is appealing, it also poses issues for
all processes involved in its design. One of the primary issues is that characterization of
these materials in different orientations is often difficult and expensive. Due to composite
materials’ anisotropy, heterogeneity, and variability, their constitutive and damage
behavior remain poorly understood.
Often due to this misunderstanding, designs that use composite materials undergo
a lengthy, difficult, and expensive procedures to produce the final product. Part of these
procedures is the finite element modeling and simulation of designed components which
requires accurate material response data. As modeling capabilities improve, provided the
proper material damage response modeling data, damage models offer the ability to predict
the damage response of designs. The ability to accurately predict damage responses in
structures is a primary contributor to a design’s development time and its overall success.
In this study, multiaxial testing via the Montana State University In-Plane Loader
was performed on two carbon fiber epoxy prepreg material systems. This testing was
performed to determine the usefulness of digital image correlation and multiaxial testing
as a means of characterizing composite materials’ damage responses and to produce data
capable of informing and validating damage models. The combination of digital image
correlation and multiaxial testing provided dense experimental results that may prove
useful to qualitatively and quantitatively inform, validate, and enhance computer finite
element modeling and analysis.
13
INTRODUCTION
Introduction to Composite Materials
Materials are a crucial part of any mechanical design. Traditional engineering
materials that are well-understood and characterized are carefully sifted through during the
design process to ensure the material meets design requirements. Design requirements
include properties such as weight-restrictions, stiffness requirements, electrical/ thermal
properties, and especially material & manufacturing costs. As design requirements are
becoming increasingly more stringent, design engineers have more difficulty determining
material systems that can meet these strict requirement combinations. Composite materials
offer a solution for complex design requirements as their properties can be catered to
specific design problems.
Of course, the introduction of a more design-flexible category of materials into the
design process is not without its drawbacks. “Composite materials” is a term that
encompasses a large range of material systems; however, this document will focus on fiber
reinforced polymers (FRPs). FRPs encompasses yet another large range of materials but
commonly known types are carbon fiber and fiberglass material systems. Aside from well-
known applications such as sporting goods or boat hulls, FRPs are also widely used in both
aerospace and wind-energy applications – both of which have been topics of research for
the Montana State University’s Composite Group and both of which are industries driving
research to better understand composites.
A better understanding of composite materials will inevitably push the boundaries
of design capabilities, increase performance, and reduce costs. As mentioned above,
14
composites can be catered to specific applications even on a part-by-part basis. However,
a design process seeking the use of composites is inherently more complex as there are
several things to consider. Rather than having a design and choosing a suitable material
that drives the design geometry, when using a composite material, the design geometry is
typically performed in tandem with the design of the material itself. So, although
composite materials may produce solutions to some problems their nature inherently
complicates the design process. As of now, composites as a whole are not well-understood
which can result in over-design, design flaws, and their use can be hinderingly expensive.
Brief History of Composite Materials
Composite materials were beginning to be used as early as the 1940s during an
important time for flight technology development. Initial instincts would indicate that
high-strength materials were suitable and appropriate for flight components such as wings
and rotorcraft blades; however, high-strength materials are typically also brittle. This
required an innovation to mix properties of both high-strength and ductile materials. Glass
fibers were prevalent during the time as they were being produced in the early 1930s by
Owens-Corning and were an ideal choice to be used as the ‘high-strength component’.
Engineers came to realize that immersing chopped glass fibers in a low-strength polymer
resulted in a stronger, stiffer, and lighter-weight polymer-like material [1].
Not long after their introduction, industry began producing and using composites
in many different fields. Still in the 1940’s the U.S. Navy took advantage of the fiberglass
reinforced polymers’ (GFRPs) electrical insulation properties for their terminal boards; and
the Wright-Patterson Air Force Base launched exploratory projects to begin using these
15
materials in structural aircraft parts. Not long after research began, in 1946 the first plane
with GFRP structural components (the Vultee BT-15) made its first flight [2]. In later years,
research and development teams have pushed composite materials farther which resulted
in high-performance materials.
In 1957, the Soviet Union successfully launched the Sputnik I satellite and spurred
on the space race. Of course, a necessity of such an endeavor is even lighter, stronger
materials than GFRPs. Asbestos fibers, phenolic resins, metal matrix composites,
carbon/graphite fibers, boron fibers, and other systems were all being developed and used
during this time as a means of solving the unique problems posed by the exit and reentrance
from a lunar mission. The space race changed everything from politics, economics, and of
course engineering problems. The National Aeronautics and Space Administration
(NASA) was founded in 1958 and NASA is still at the forefront of not only space travel
but engineering innovations such as composite materials and their applications.
Since the space race and space era, the industrialization and commercialization of
composite materials has grown immensely. Now, composite materials can be found in
nearly everything, from consumer products like golf clubs and tennis rackets to passenger
airplanes and spacefaring devices & vehicles. Again, industries like aerospace are pushing
research and development to better understand composite materials so that their use in
design is appropriate. A better understanding of the materials themselves can assure higher
performance, less expensive, and safer designs. To better understand materials, their
properties must be measured and understood – not only engineering properties such as
moduli and density but their damage characteristics as well.
16
Composite Material Characterization
It is prudent to define “composite materials” further before continuing. It has
already been stated that this document will focus on FRPs which already suggests a mixture
somehow between fibers and polymers. A generally accepted definition of “composite
material” is a combination of two or more distinct materials to form a new material with
enhanced properties [3]. It is also important, from a fundamental standpoint, to know that
the combination of distinct materials mentioned above is on the macro-scale and not the
micro-scale. This means that the mixture of materials is the mixture of ‘larger’ structures
and not on a molecular level; typically, alloys are not considered composites. Composites
are non-homogeneous (heterogeneous) materials.
Knowing this, it can be easily seen how complex the characterization of these
materials can be. Traditional engineering materials such as metals have been studied and
characterized in depth, some of which for thousands of years. Only in the past
approximately 70 years have composites been studied, and herein lies the first difficulty.
It is difficult to understand much about a material that is still in its infancy, they have simply
not existed long enough. However, the youth of the material isn’t even the largest concern.
Of course, just because the material is relatively new compared to traditional
engineering materials (e.g. metals) doesn’t mean that previous gained knowledge, methods,
and technology do not apply to it. However, when applying knowledge such as mechanics
of materials and fracture mechanics to metals, it is seemingly infinitely easier than doing
so for composites. Metals are expected to behave a certain way under given conditions
and variations can be intuitively, analytically, and experimentally rectified. Composites,
on the other hand, are not so simple nor easily understood; they pose unique difficulties in
17
each of those categories: intuitively, analytically, and experimentally. As such, now and
over the past 70 years, the effort to better understand composites is a necessity for their use
in many design situations.
Materials are characterized by performing several series of tests to obtain pertinent
material information. This characterization of the material is then used in the design
process for material selection. Typically, when well-understood materials e.g. steels or
aluminum alloys are used in a design, the design process is fast and the properties obtained
from previous tests applies directly to the final manufactured part. This allows the designer
to use analytical and experimental methods to validate the design. However, the use of
composites as mentioned above, is much more complex as the material is typically
designed in tandem with part geometry. Since the bulk material is being designed with the
part, the design process is much slower and far more expensive. MIL-HDBK-17 is a
guidance document prepared for departments and agencies of the U.S. Department of
Defense which eloquently states:
“Analysis alone is generally not considered adequate for substantiation of
composite structural designs. Instead, the ‘building-block approach’ to
design development testing is used in concert with analysis. This approach
is often considered essential to the qualification/certification of composite
structures due to the sensitivity of composites to out-of-plane loads, the
multiplicity of composite failure modes and the lack of standard analytical
methods.” [4]
18
Figure 1. The "building-block approach" schematic as shown in MIL-HDBK-17-1F [4]
The schematic from Figure 1 (above) depicts the complexity of designing especially
large components using composites. Not only are tests necessary on the coupon level, the
design cannot be validated without substantially more tests up to and including the final
component. As knowledge of composite materials is increased, more data from the least
expensive (coupons) level can be used while mitigating the expense of component and sub-
component tests to validate designs. Furthermore, MIL-HDBK-17 also mentions the
complexity of damage in composites as a hindrance to design validation.
19
Damage and Failure in Composite Materials
During WWII and the space era, a need to better understand material failure drove
research further. These research efforts were accelerated during the time that passenger
planes, such as the de Havilland Comet, became a growing market and necessitated a better
understanding of fatigue-induced damage [5]. Out of the research efforts to understand
damage, academic areas like fracture mechanics emerged as a means of explaining material
damage and failure. Fracture resistance is typically measured in terms of a fracture
toughness value (𝐾) or a strain-energy release rate (𝐺𝐶) for a material. Both ideas originate
from the same principles and result in accurate methods of determining damage and failure
analytically for traditional homogeneous engineering materials such as steels. However,
when applied to composites, these methods become multiplicatively more difficult to apply
due to the nature of composite materials.
The heterogeneous nature of composite materials inherently complicates how
damage initiates and propagates through the material. Damage is no longer characterized
simply as a crack that grows but can consist of many mechanisms. Damage can initiate
and progress through a minimum of three unique mediums of the material – the fibers,
matrix, and fiber-matrix interface. Figure 2 below depicts some intra-ply damage
mechanisms that can occur within a lamina. Already, within only one ply, there are many
types of damage that can occur within a composite material during the damage process.
However, this does not account for damage that can occur through the thickness of the
laminate or inter-ply damage such as delamination of neighboring plies. Although typically
not all of these damage modes occur at the same time, their mere existence complicates
efforts to understand, model, analyze, and even evaluate damage in composite materials.
20
Figure 2. Intra-ply damages in a composite material (by Anderson [6]). (1) Fiber Pull-out.
(2) Fiber Bridging. (3) Debonding. (4) Fiber Breakage / Rupture. (5) Matrix Plasticity &
Cracking
Even while considering a composite’s heterogeneity and some of its damage
mechanisms, this simply does not and cannot account for all variables that affect the
damage of a composite material. More variables affecting composite materials’ damage
characteristics include topics such as fiber misalignment & waviness, matrix porosity,
creep behavior, moisture content, and many other defects and factors. This complexity
necessitates more general methods of investigating damage in composites rather than
attempting to completely understand each contributing factor fully; some of these methods
include a variety of statistical analyses, energy methods, and even educated guess & check.
It is important to note that damage is not necessarily synonymous with failure and
that failure is defined by the application of the material. When a structural member no
21
longer has the capacity to perform its intended design that member is said to be failed; this
allows, in most designs, a degree of damage to accumulate in a component before failure.
Certainly, damage and failure are related, but their relation is determined by the design
engineer. This further demonstrates the necessity of understanding damage in composite
materials – at some point, damage accumulation will cause failure and a better
understanding of damage can not only lead to the mitigation of damage by means of better
designs but also better damage-tolerant designs*.
Motivation
After being introduced to the complexities of composite materials, it is evident that
a better understanding of composites is needed to streamline the design process and
produce better designs. Typically, in the design process, finite element models are used to
simulate design responses to its application loads and displacements. This mitigates the
need for building expensive prototypes and test objects for the validation of a design. Finite
element models are improving to the point of predicting damage but are still generally
unrepresentative due to the lack of proper damage characterization of the materials and the
complexity of the models. This study aims to produce data capable of informing and
validating these damage models so that the development and application of damage models
is more representative of the final design. If damage models can more accurately predict
damage, their use in the design process may reduce design time, design expenses, and
ultimately improve the final design.
* Damage Tolerant Designs account for development of damage primarily due to fatigue and ensure, using
specific inspection intervals, safe operation up to at least the next appointed inspection interval.
22
THE MONTANA STATE UNIVERISTY IN-PLANE LOADER
The Montana State University In-Plane Loader will be referred to as the “IPL” from
this point on. There are many previous theses (to be discussed further later) that sum up
the previous works with the IPL. However, for the sake of completeness and to provide
information on the current state of the IPL, this chapter will introduce the purpose,
conception, previous work, recent changes, and current work with the IPL.
Multiaxial Testing
This section will discuss, in brief, why multiaxial testing can be a powerful asset in
the understanding and characterization of composite materials and will ultimately discuss
the purpose of the IPL. As mentioned above, materials undergo several series of tests to
determine pertinent engineering properties necessary for informed and efficient design.
For isotropic materials†, these tests are not direction-dependent and can therefore be
characterized fully without having to account for material orientation. The anisotropy and
heterogeneity of composite materials dictates that their characterization undergoes a much
more rigorous process accounting for properties in all relevant directions. This can be seen
by considering even a single unidirectional ply (or lamina) and even more so with materials
involving stacks of plies (or laminates).
Traditional testing such as those performed with uniaxial testing machines can
supply information typically in a single direction by means of uniaxial testing. In contrast
† Isotropic materials – Materials where the elastic properties are the same in all directions [10]. Traditional
engineering materials such as metals are often idealized and treated as isotropic materials.
23
to composites, isotropic materials have infinitely many planes of material property
symmetry, allowing properties to be applied to all directions of the material where few tests
can fully characterize the material. This not only applies to properties such as moduli
(𝐸 𝑜𝑟 𝐺), but also failure properties such as fracture toughness (𝐾) as mentioned above.
However, these uniaxial tests are not sufficient to fully characterize composite materials’
stiffness properties and especially failure/strength properties. As a means of justifying the
use of multiaxial testing, the examination of both elastic theory and failure theory can be
examined individually. However, for the sake of brevity, only failure theory will be
explained in depth as it pertains more to this thesis. Before continuing though, it is
necessary to explain, on a basic level, elasticity of anisotropic materials.
{휀} = [𝑆]{𝜎} (1)
In general, Hooke’s Law states that stress is proportional to strain based on the
elastic properties of the material as shown in Equation (1) above. This relationship is
typically expressed more simply for isotropic materials where only three elastic constants
(𝐸, 𝜈, 𝐺) are needed to fully describe the material’s elasticity. However, for an anisotropic
material with no symmetry whatsoever, the compliance matrix [𝑆] is expressed as a matrix
with 81 independent terms. This is then reduced, using material symmetries and other
assumptions to produce a more manageable relationship. This resolution for composite
laminae is generally expressed in its simplest form as Equation (2) below, where
assumptions such as transverse isotropy and plane stress are applied [3].
{
휀1
휀2
𝛾6
} = [
1/𝐸1 −𝜈21/𝐸2 0−𝜈12/𝐸1 1/𝐸2 0
0 0 1/𝐺12
] {
𝜎1
𝜎2
𝜎6
} (2)
24
From Equation (2), there exist different properties in different directions for
anisotropic materials where subscripts 1, 2, and 6 (also expressed as 12) correspond to the
material local coordinates fiber-direction, transverse-direction, and shear on the 1-2 plane
respectively. This notation will be used in the following section to express failure criterion.
Failure and Strength of Composites
As discussed above and shown in Figure 2, failure in composites is complex and
for this reason, there exists an area dedicated to understanding failure/strength of
composites. Generally, there exists three types of failure criterion applied to anisotropic
materials namely composites: Maximum Stress Criterion, Maximum Strain Criterion, and
Interacting Criteria, the latter of which justifying the use of multiaxial testing. Each of
these criteria attempt to predict failure of composites by a means of simple calculations for
use during design and analysis. Failure criteria attempt to calculate the limit of a material
before its mechanical properties begin to change. These criteria do not attempt to account
for physical damage to the material such as microscopic cracking etc. until the coalescence
of this physical damage results in material property change (mechanical damage).
It is important to note that the failure criterion introduced in this section have been
developed from mature and well-understood principles in mechanics. However, no matter
how mature, any failure criterion (even criterion not explicitly stated in this thesis)
developed to date has its own pros and cons. This can be demonstrated by the World-Wide
Failure Exercises (I & II) [7].
The exercise was the work of several experts to determine the accuracy and validity
of 12 different failure criteria. Christensen [8] summarizes the work of the exercise well
25
in stating that there are very large variabilities in both the failure criteria and the material
samples. Many of these failure criteria differ greatly from one another but, for different
instances, well-represent experimental data. The true outcome of the exercise was that
there exist no reliable failure criteria for fiber reinforced composite materials that are
consistently accurate.
Maximum Stress Criterion
Maximum Stress Criterion. Maximum Stress Criterion is expressed by a series of
inequalities defining limits on the stress of the material. These strength limits are defined
by experiments in both the fiber or 1 direction, and the transverse or 2 direction for both
tension and compression cases as well as for shear or 6 deformation.
𝜎1 > 𝐹1𝑡 |𝜎1| > 𝐹1𝑐
𝜎2 > 𝐹2𝑡
|𝜎2| > 𝐹2𝑐
|𝜎6| > 𝐹6
𝑖𝑓 𝜎1 > 0𝑖𝑓 𝜎1 < 0𝑖𝑓 𝜎2 > 0𝑖𝑓 𝜎2 < 0
(3)
The inequality expressions are shown in Equation (3) above and state that failure
of a lamina is predicted when at least one of the stress (𝜎) in the material coordinates
exceeds the corresponding experimental value of strength (𝐹) [3]. Also, as stated by
Barbero [3], this criterion proves useful in that it gives information regarding the mode of
failure. However, a drawback of this criterion is that it does not provide interaction
between stress states and therefore is not conservative for stress states that are not
dominated by one component of stress. This criterion is most effective where a single
component of stress dominates the stress state of the material.
Maximum Strain Criterion
26
Maximum Strain Criterion. Maximum Strain Criterion is similar to Maximum
Stress Criterion except the limits are no longer stress values but strain values (strain to
failure). Again, these limits are defined by experiments in both the fiber direction (1) and
the transverse direction (2) for both tension and compression cases as well as for shear (6).
This criterion uses the concept of strength ratios and the criterion is shown in Equation (4)
below.
𝑅1 = 휀1𝑡/휀1 𝑅1 = −휀1𝑐/휀1
𝑅2 = 휀2𝑡/휀2
𝑅2 = −휀2𝑐/휀2
𝑅6 = 𝛾6𝑢/|휀6|
𝑖𝑓 휀1 > 0𝑖𝑓 휀1 < 0𝑖𝑓 휀2 > 0𝑖𝑓 휀2 < 0
(4)
This criterion states that failure is predicted when at least one of the strain values
reaches the experimental strain limit values or when a strength ratio (𝑅) meets or exceeds
a value of 1. This criterion is the most popular failure criterion used in industry today per
Barbero [3]. This is likely due to the ability to directly measure strain in tests unlike the
Maximum Stress Criterion where stress must be computed and, for this reason, may be less
reliable. An inherent usefulness of this criterion is that Poisson Effects are included –
something Maximum Stress Criterion fails to account for. Also, like Maximum Stress
Criterion, Maximum Strain Criterion provides information regarding the mode of failure.
Interacting Criterion
Interacting Criterion. There are several different well-accepted models that fall
under interacting criteria as a category. Tsai-Wu is a popular criterion heavily used for
polymer matrix composites. Tsai-Hill is often applied to special cases of FRP analysis or
materials like ceramic and metal matrix composites. The basis behind all interacting
criteria is that they account for the interaction between different failure modes which the
27
previously mentioned criteria fail to include. This is more realistic as many modes of
failure are not physically justified to be independent from one another. However, to be
addressed later, quadratic failure criteria such as Tsai-Wu and Tsai-Hill also pose problems
of their own.
The previously discussed criteria (Maximum Stress & Strain) assume that a
multiaxial state of stress can be represented as a series of uniaxial states that are
independent from one another. If any of the uncoupled, uniaxial states exceed their
corresponding strength values, then the material is determined to have failed. For a
multiaxial case, the fact that the material is being subjected to combined stresses is ignored.
So, when two or more failure modes interact to produce failure, the Max Stress & Strain
criteria may result in unrepresentative predictions and ultimately unconservative or over-
conservative designs. These interactions can be intuitively seen and further demonstrated
by looking at some cases that Barbero [3] describes:
• Transverse tensile 𝜎2 > 0 and shear stress 𝜎6. In this case, both stress states
can produce the same type of damage (matrix cracking) and therefore
interact with one another detrimentally.
• Transverse compressive 𝜎2 > 0 and shear stress 𝜎6. In this case, interaction
is opposite from the previous. Rather than negatively affecting the material,
these states of stress oppose one another. Cracks in the matrix produced by
shear must overcome the transverse compression stress to propagate in the
material.
28
• Longitudinal compressive 𝜎1 < 0 and shear stress 𝜎6. This case can be
detrimental to the material as the shear stress can produce damage and
effectively reduce shear modulus 𝐺12 which, in turn, reduces the
longitudinal compressive strength.
Even though interacting criteria attempt to reconcile this interaction between
stresses, there are drawbacks to this category of failure criteria [3]. First, many interacting
criteria, including Tsai-Wu and Tsai-Hill, force an artificial interaction between matrix and
fiber modes of failure. Second, interacting criteria usually do not provide information
regarding the mode of failure (matrix or fiber). An example of an interacting criteria, posed
by Barbero, is displayed in Equation (5) below.
⟨𝜎1𝑓
𝐹1𝑡⁄ ⟩ + ⟨−𝜎1𝑓
𝐹1𝑐⁄ ⟩ − 1 = 0
𝑓22(𝜎2𝑓
)2
+ 𝑓44(𝜎4𝑓
)2
+ 𝑓55(𝜎5𝑓
)2
+ 𝑓66(𝜎6𝑓
)2
+ 𝑓2(𝜎2𝑓
)2
− 1 = 0
𝑓2 =1
𝐹2𝑡+
1
𝐹2𝑐
𝑓22 =1
𝐹2𝑡𝐹2𝑐 ; 𝑓66 =
1
(𝐹6)2
𝑓44 =1
(𝐹4)2 ; 𝑓55 =1
(𝐹5)2
(5)
Finally, as a result of examining failure criteria, it is evident that the world of
composites is in need of a well-informed and reliable failure criterion for fiber reinforced
materials. Also, to obtain this criterion, the interaction between stresses must be
considered. These interactions must be validated and informed by experimental data and
herein lies the purpose and endorsement of multiaxial testing.
29
Brief History of the IPL
Inspired by work from the Naval Research Laboratory (NRL) [9], the IPL was
conceived in 2001 as a senior design project for mechanical engineering undergraduates
Eric Booth, Ken Higgins, and Marc Schaff [10]. The first-generation IPL was then fully
completed later by several graduate and undergraduate students. Since the first-generation,
the IPL has undergone many renovations and been the primary topic of several Master’s
theses. Each generation of the IPL has improved the hardware, software, and
instrumentation reliability of the machine.
Figure 3. Schematic of possible deformations applied to a coupon in the IPL
The first-generation IPL was laid horizontal on a table as seen in Figure 4. The
design utilizes three strategically placed stepper motor-driven ball-screw actuators to
deliver displacements and loads to the sample. The actuator configuration allows two
30
translational components (𝑑𝑥 and 𝑑𝑦) and a rotational component (𝑑𝜃) shown above in
Figure 3 (this will be discussed further later) to be applied to the sample without the
machine binding against itself. Pancake load cells were installed at the end of each actuator
to accurately measure each supplied load. Instrumentation and control of the IPL was
performed in LabVIEW, Mathematica, and MATLAB software packages [11]. As stated
above, this original design has been upgraded several times resulting in the current ‘fifth-
generation IPL’ shown in Figure 5.
Figure 4. A photograph of the first-generation IPL as shown in Ritter's thesis [12]
Third Generation
The third-generation IPL contributed primarily to Smith’s thesis [13].
Improvements made from the first and second generation IPL in this iteration were drastic.
31
Both Smith and Schmitt [14] summarize the changes made to the IPL in great depth. For
completeness, the primary contributions to this generation of IPL are listed below.
• LVDT Position Control – The first-generation IPL relied on encoders in each
actuator to control position of the cross-head. This was then changed to be
controlled by LVDT’s mounted directly to the grips. Schmitt [14] describes the
LVDT array used.
• New Gripping Mechanism – The first-generation IPL used knurled grip plates
mounted in an assembly which required bolts to clamp the sample. This grip
assembly was then replaced by an assembly using hydraulic pistons driving self-
centering carbide-textured grip plates to clamp samples in place.
• Upright Mounting – The horizontal configuration of the first-generation IPL proved
to be cumbersome when using the new gripping mechanism. Thus, a stand was
built to hold the IPL upright so that each side of the grips could be easily accessed.
• Out-of-Plane Constrainer – A bearing plate was mounted directly to one end of each
grip half to reduce out-of-plane response during testing. Out-of-plane response
occurred when loading a sample due to compliance of the machine itself and off-
axis loading of the coupon. This addition was an attempt to mitigate any
contribution of out-of-plane displacements applied to the coupon.
This concludes the primary changes made since the first-generation IPL. However, see
previous works for a more complete list of changes and Collett [11] for a validation study
of the third-generation IPL.
32
Fourth Generation
As stated in Schmitt’s thesis [14], the fourth-generation IPL did not undergo any
large changes to mechanical components or software of the system but only the addition of
a digital camera to monitor the distortion of the coupon mid-test. This camera system, run
via MATLAB, is a digital image correlation (DIC) system designed and implemented by
Parker [15]. Since Parker’s DIC implementation in 2009, Montana State University has
purchased a commercial DIC system to replace Parker’s. The commercial DIC used for
this research is called GOM ARAMIS. The purpose and application of DIC as it pertains
to this research will be thoroughly explained later; for technical inquiries about DIC and
its development, refer to Parker [15].
Fifth Generation
The most current generation of the IPL utilizes most of the same components as
shown in Figure 4. The ball-screw actuators, pancake load cells, and frame have been
changed little since the original design. Even as such, the most current fifth-generation
IPL shown in Figure 5 has been the largest redesign of the IPL. Many of the
recommendations from previous research were taken into account to improve the reliability
and performance of the IPL. The primary concerns and design goals addressed in the
redesign were:
33
Figure 5. Fifth-generation Montana State University In-Plane Loader. A: Out-of-plane
Constrainers (both sides) B: Coupon Loading Location C: Main Electronics
• Coupon Slippage – Slippage of the coupon in the grips during testing (also refered
to as grip slippage) remained a problem even after the implementation of the third-
generation grip assembly.
• Out-of-plane displacement – Again, even though the out-of-plane constrainer was
employed to mitigate this problem, out-of-plane displacement remained an issue
during testing. Out-of-plane displacements typically occurred due to high loads
especially during damage of the coupon.
• Software Issues – The third-generation IPL employed new position-control code
run in LabVIEW that ultimately caused issues due to noise in the LVDT readings
and control code feedback loop inefficiencies. The most noteable effect of these
34
inefficiencies was inconsistent loading during testing (rapid, jerky displacements
provided by the actuators).
• GOM ARAMIS Implementation – The implementation of GOM ARAMIS DIC as
the primary data acquisition system for both loads and displacment data during
testing. Ideally, optical measurement systems such as DIC should be used to control
the IPL (see recommendations), but DIC at this point was implemented as data
acquisition only.
The resolution of these issues required several hardware changes to be made on the IPL
including a new grip assembly, new out-of-plane constrainers, and new electronics.
Included with these hardware changes, an entirely new version of IPL control code was
written. These changes and a description of the current state of the IPL are discussed below.
A solid model of the fifth generation grip assembly can be seen in Figure 6. The
latest grip design was made to grip the coupon not only on the faces but also the edges
using the transverse support plates (‘C’ in Figure 6). These transverse support plates supply
little, if any, load to the sides of the coupons and are meant only to restrict coupon
movement in the grips during rotations and x-displacements. The grips are still driven
closed using the same hydraulic piston (A in Figure 6) supplying a load up to 7800 𝑙𝑏𝑓 to
the coupon. The new grip plates (‘B’ in Figure 6) have the same carbide coating applied
to the surface but now over a larger effective grip area (2 𝑖𝑛 × 2 𝑖𝑛). The new grips no
longer have a self-centering mechanism and must be shimmed to maintain in-plane loading.
Finally, these grips can support coupons up to 0.5 𝑖𝑛 thick and 2 𝑖𝑛 wide which is an
35
improvement over previous grips. The thinnest allowable coupon with the use of the
transverse support plates is 0.125 𝑖𝑛.
Figure 6. Solid model rendering of the fifth-generation IPL grip assembly. A: Hydraulic
Piston B: Carbide-textured Grip Plates C: Transverse Support Plates
The new out-of-plane constrainers are shown as ‘A’ in Figure 5. Each of the 4 out-
of-plane constrainers is made from 2 slewing bearings, 2 linear bearing carriages, and a
linear bearing rail. A slewing bearing is mounted directly to the frame (both the base and
the cross-head) with a linear bearing carriage mounted directly on top of it. The linear
bearing rail then connects the upper and lower assembly and is only allowed to slide in the
top assembly. The slewing/ linear bearing configuration allows for translational and
rotational displacements in-plane but restricts any out-of-plane displacements of the cross-
head. For future modifications, the linear bearing rails were designed so that supports may
be mounted to them to supply more rigidity if necessary.
36
The latest electronic hardware is shown as ‘C’ in Figure 5. These new components
consist of an up-to-date data acquisition (DAQ) module and actuator control module.
These components bridge the connection between the latest IPL control software and its
electrical components. The DAQ serves to gather load cell and LVDT voltages as well as
supplying voltages (voltage output) for resolved loads to the ARAMIS DIC. The actuator
control module serves as the interface between LabVIEW software and the actuator motor
drives.
The latest IPL control software is written using the same basic control scheme by
means of a kinematic vector loop solution. The original vector schematic has been updated
to follow the schematic shown below in Figure 7 where vector ‘k’ is computed and
controlled by changing actuator lengths ‘𝐿1’, ‘𝐿2’, and ‘𝐿3’. Figure 7 depicts vectors that
can be controlled as blue and uncontrollable vectors as red. Vector ‘𝑘’ is the gap between
the bottom and top grips and controlling this displacement controls the deformation of a
sample when loaded in the grips
37
Figure 7. Latest vector loop schematic and nomenclature as used for IPL control software
Aside from a more efficient control scheme, there are two notable changes made to
the IPL. First, the LVDT array for position control of the crosshead has been removed and
the LVDTs have been repurposed as calibration aids for accurately determining ‘𝐿1’, ‘𝐿2’,
and ‘𝐿3’ at any given time. The position control is now based on encoder readouts in the
actuators rather than using the LVDT array. This decision was made for several reasons
but was mostly due to LVDT noise errors that came from measuring miniscule
displacements close to the grips. Second, the new DAQ allowed for resolved load output
as voltages to the ARAMIS. This allows the ARAMIS software to be used for all data
acquisition from the IPL including optical displacement measurements and load data.
38
Application of the IPL
Dissipated Energy Density
Introduced with the idea of an In-Plane Loader, dissipated energy density was
formally introduced by the NRL in 1995 as a means of quantifying damage in composite
materials undergoing multiaxial loading conditions. The full scheme for using dissipated
energy density to quantify damage can be seen in the NRL’s published paper [9] Ritter’s
thesis [12], and Schmitt’s thesis [14]. The basics of the scheme and the application as it
pertains to this research are discussed below.
Figure 8. A generic load vs displacement plot as seen in Collett [11]. Only to illustrate
dissipated energy
39
As a material is loaded, there typically exists a linear-elastic response until
softening occurs. After softening occurs, the force vs. displacement curve levels off slowly
as energy is dissipated. Finally, once the material is failed entirely, the load drops to zero.
Ideally, if the material is unloaded at any point during the test the load and displacement
will return to zero in a linear fashion – this portion of energy under the curve (bounded by
the linear unloading line) is referred to as recoverable elastic energy. The portion of the
curve that is not recovered is referred to as the dissipated energy. A graphical representation
of this can be seen above in Figure 8. For a composite material, the softening phenomenon
discussed above is idealized as contributing entirely to damage within the material.
Although this assumption is not entirely accurate due to some typical non-linear elastic
response in composite materials, it is a conservative assumption and will be used.
The idea proposed by the NRL was to use calculated dissipated energies at many
points within a multiaxial test to determine the damage state of the material. Without too
much detail, as in Schmitt [14], the basic steps of using dissipated energy density as a
metric for damage is as follows:
1. Multiaxial mechanical testing in the In-Plane Loader
2. Measure boundary displacements and loads from the test
3. Measure experimental dissipated energies at many points throughout the test
4. Recreate test using a finite element model and obtain analytical dissipated energies
5. Minimize the difference between the analytical dissipated energy from step 4 and
the experimental dissipated energies from step 3 by controlling the model
40
6. Once this difference is mitigated, use the finite element model strain values to
determine “actual dissipated energy function”
7. This “actual dissipated energy function” is defined completely by in-plane strains
and is proposed to act as a material property for quantifying damage.
8. This “actual dissipated energy function” can then be applied to individual elements
in a model as a means of creating a functional continuum damage model.
a. Once critical dissipated energy within an element is reached (based on
strains), this element is said to have failed and its properties are altered or
the element is broken.
This was a brief outline of the scheme. For more information regarding previous work
using this dissipated energy scheme as a metric for damage see NRL’s publication [9].
For several reasons, it was decided for this research to not use dissipated energy
density as a metric for quantifying damage during the test. After many publications and
research trying to produce accurate progressive damage models using this method, there
has never been a well-accepted model produced. This may be due to the “circular logic”
of the scheme itself. In a way, a model is created to define damage properties for itself.
Boundary conditions and measured dissipated energy are obtained from actual tests,
assumed to be accurate, then used to adapt an idealized model. This produces an equation
used to control the identical model with hopes of virtually reproducing the experimental
test. In a way, this is very accurate but not a rule that generalizes well to many damage
modes and states as shown by the work of Schmitt [14] and Smith [13]. Without
normalization, these dissipated energy density calculations proved ineffective. This is not
41
stating that this method does not have merits, only that this research will not use dissipated
energy the same way.
As a note, dissipated energy density is not the same as total dissipated energy.
Dissipated energy density can be thought of and treated as an extension of strain energy
release rate (a fracture mechanics material property). Dissipated energy cannot be treated
as a material property but is useful as explained further below.
Dissipated energy as it pertains to this research is much simpler and, in a way, more
primitive than dissipated energy density. For each sample, total dissipated energy is
calculated only to detect the onset of damage and is never used as a metric to quantify
damage accumulation. This will be explained further later; but as an example of a total
dissipated energy calculation, see Figure 9 below. The dissipated energy is calculated for
each deformation component (x-direction, y-direction, and rotation) independently then
summed for “total dissipated energy”.
42
Figure 9. Example dissipated energy calculation for sample 11_008. Dissipated energy is
expressed in units of 𝑙𝑏𝑓 ∙ 𝑖𝑛
It can be easily seen that the dissipated energy ‘jumps’ at “stage 62”. The “stage”
axis is the image step for the test. At stage 62, there is in increase in dissipated energy for
each deformation component (𝑥, 𝑦 and 𝜃). Again, assuming all dissipated energy is
attributed to damage in the material, the conclusion of the Figure 9 is that damage initiation
occurs at stage 62 of the test and continues to progress until ultimate failure at stage 105.
The calculation process, purpose, and place in post-processing will be described more fully
later.
43
It is important to note that the method behind this total dissipated energy calculation
is the same as shown in previous works. However, even though the calculations are
handled the same way, the input displacements are now far more accurate as they are
measured directly from the digital image correlation data rather than an assumed
displacement taken from the IPL. The primary risk of using the assumed displacement data
taken at the grips of the IPL is that it does not account any coupon slippage and assumes a
perfect gripping scenario. Using new software and methodology for measuring
displacement, the calculation of dissipated energy is more accurate.
Digital Image Correlation
Strain gauges and extensometers are a well-accepted industry standard for
measuring strains in quasi-static mechanical tests. However, these two methods of strain
measurement are discrete measurements that average strain values over the entire applied
area of the test coupon. Therefore, localized strains (at stress concentrations and damage
initiation sites) are nearly impossible to capture during a test using these methods. Strain
information at stress concentrations and damage initiation sites is grossly determined via
analytical methods with applied assumptions. Assuming material properties, including
damage characteristics, often produces error since damage characteristics and even
constitutive responses of composite materials are not fully understood (this is the data being
tested for). So, like other challenges of material testing, this necessitated different
methodology in measuring strains.
There are many types of full-field strain measurement methods. However as
mentioned previously, due to Parker’s work on digital image correlation, a commercial
44
DIC system (GOM ARAMIS) was purchased to satisfy the need for full-field strain
measurement. The use of DIC doesn’t only provide full-field strain data but is useful (as
seen above) to measure boundary displacements, correlate to finite element models, present
data, and much more.
The idea of digital image correlation is simple yet the application of it is difficult.
For inquiries about the theory and technicalities of DIC refer to Parker [15] and Hild [16].
The basic premise of DIC is that through a series of non-contact optical measurements and
analysis of these images, deformations can be calculated. To do this, DIC recognizes the
surface structure of the measured object in digital camera images and allocates coordinates
to the images’ pixels. As the measured object is deformed, the software can keep track of
the surface structure and calculate displacements based on the new images’ coordinates.
More simply, DIC software tracks the displacements of each speckle on a pixel-level from
image to image to calculate the full-field deformation. An example of the stochastic‡ spray
pattern used for DIC can be seen in Figure 10 below.
‡ Stochastic – A randomly distributed pattern that may be analyzed statistically but not necessarily
predicted.
45
Figure 10. Image of acceptable stochastic pattern with 15×15 facets [17]
The facets represented in the image above are analogous to elements in a finite
element model. The smaller the facet, the higher resolution deformation calculations will
be but at the cost of increased computation time. Facet boundaries are defined based on
unique high-contrast areas of the spray pattern. These areas that define the facet boundary
are what the software tracks throughout each image. Deformation of the measuring object,
or even a view from another camera, results in a change within the calculated facets as
shown in Figure 11 below. For this research, facets with dimensions 15×19 pixels are
used to maintain a manageable computation time but have reasonable resolution.
Figure 11. Facet tracking as shown in the ARAMIS manual [17]
46
An example of a test that has been fully post-processed within ARAMIS can be
seen in Figure 12 below. In Figure 12, there are four images displayed. The left-most
image shows displacement vectors for each calculated facet. The second, third, and fourth
images show computed strains displayed as gradient overlays (휀𝑥, 휀𝑦 , 휀𝑥𝑦 respectively).
There is also additional information shown on the graphs and in the table including load
data, frame time, and test rate.
Figure 12. Example of ARAMIS post-processed data and data presentation. (Still image of
a video taken just after damage initiation)
The ARAMIS hardware system is composed of a computer, sensor control module
(DAQ), and the sensors (cameras). The image shown in Figure 13 below shows the
computer and cameras. It is important to note that the cameras are a stereo system, meaning
47
they observe the measuring object from different angles and allow the calculation of depth
(z-axis) deformation. This stereo setup is beneficial for two reasons. First, correlation
between the left and right camera images for each stage produces much more accurate
results than a single-camera setup. For this setup, average measurement error is
consistently computed to be less than 0.00003 𝑖𝑛. Second, as mentioned, these two
cameras allow for the calculation of depth (z-axis) which assures that in-plane tests are in
fact in-plane. In practice, local coordinate systems can be created for complex part
geometry using these features as well.
Figure 13. GOM ARAMIS hardware as shown in ARAMIS Manual [17]
48
Test Coupon
Material Systems
Both material systems explored are carbon fiber reinforced polymers (CFRPs).
They are unidirectional prepreg§ systems composed of a single continuous carbon fiber
type and a single resin system. This prepreg system allows for the highest industry standard
fiber volume fraction, highest quality fiber orientation, and even complex shapes to be
produced. The first material system to be explored is a Hexcel 8552 IM7 (also referred to
as IM7/8552). The second system is a more brittle and slightly thicker material system
produced by Toray. Material properties pertinent to this work are displayed below in Table
1. These material properties will be used to produce a finite element model.
Table 1. Lamina material properties for both material systems. Material properties are
specified for material at room temperature and dry (RTD) [18]
Material Property IM7/8552 Value Toray Value
E1 (msi) 23.51 20.6
E2 (msi) 1.30 1.13
E3 (msi) 1.30 1.13
12 0.356 0.34
13 0.356 0.34
23 0.53 0.53
G12 (msi) 0.68 0.58
G13 (msi) 0.68 0.58
G23 (msi) 0.38 0.37
Furthermore, and useful later, is a list of strength properties for IM7/8552 shown
below in Table 2. All values for IM7/8552 are from the National Institute for Aviation
§ Prepreg – a fibrous material preimpregnated with a resin system. The resin system is typically a partially
cured epoxy already containing the appropriate curing agent.
49
Research (NIAR) [18], and all values for the Toray properties were provided. No strength
properties were provided for the Toray material system.
Table 2. Strength properties for IM7/8552. Properties are specified for material under
quasistatic loading at room temperature and dry (RTD) [18]
Strength Property IM7/8552 Value (ksi)
F1_tu 371.08
F2_tu 9.29
F1_cu 251.13
F2_cu 41.44
F12 13.22
Several different laminates were provided for each of these material systems. For
this research, three different laminates of each material system were tested and analyzed.
Stacking sequences and naming convention numbers are displayed in Table 3 below. Each
sample is named using the format XX_0YY where the “Layup Number” from Table 3
replaces XX and YY is replaced by the appropriate “Test Number”. For example, the name
11_001 would be the first test performed on the [-45/90/45/0]s layup of the IM7/8552
material system. These numbers are for bookkeeping only and have no significance other
than to keep track of tests. For a full list of tests, refer to the test matrices in the appendix.
As a note, the stacking sequences are displayed “as-tested” and are not the stacking
sequences “as-manufactured”. For example, the stacking sequence [-45/90/45/0]s is
actually manufactured as [45/90/-45/0]s but measurements were taken from the ‘back side’
of the coupon. Measurements were decidedly taken from this side of the coupons since the
‘back side’ of the material was smoother than the ‘front’ and was a better surface for use
with the DIC.
50
Table 3. Laminate stacking sequence table and layup numbers. Naming convention of
tests uses "Layup Number" to replace XX of XX_0YY
Stacking Sequence Layup Number
IM7/8552 Toray
[-45/90/45/0]s 11 1
[0/90/0/90]s 13 3
[-45/45/-45/45]s 14 4
Geometry
After a compact tension sample study conducted by Ritter [12] in 2005, the
“standard” IPL coupon was created to resemble a ‘modified compact tension sample’. The
most current version of the IPL sample is shown below in Figure 14. The purpose of the
notch (also referred to as a strain riser) is to create a stress concentration in the sample as
an attempt to force failure to the middle of the gauge section of the coupon rather than near
the gauge section boundaries. This geometry was chosen as it produces the desired failure
location and proved to be easily reproducible from a manufacturing standpoint.
51
Figure 14. Latest coupon geometry. Identical gauge section to previous samples but larger
grip areas. The displayed coordinate system is not displayed at the working origin.
Reference below, Figure 20
Manufacturing
Large flat panels were received for each of the above specified material systems.
The panels measured 24 𝑖𝑛 × 48 𝑖𝑛 and needed to be cut down to dimensions shown in
Figure 14 for testing. This process took several steps to produce the coupon geometry and
is summarized below:
52
1. The received 24 𝑖𝑛 × 48 𝑖𝑛 panel was cut down into minimum nominal 10.5 𝑖𝑛
strips while maintaining and tracking material orientation. This step was only to
make manufacturing more manageable.
2. Each 10.5 𝑖𝑛 strip produced in step 1 was then cut in half lengthwise to produce
material strips with the proper 5.125 𝑖𝑛 sample height.
3. Strips produced in step 2 were then cut slightly larger than the specified 1.00 𝑖𝑛
width shown in Figure 14. This produced samples 5.125 𝑖𝑛 tall and just over
1.00 𝑖𝑛 wide.
4. The edges of the samples produced in step 3 were then sanded carefully to remove
edge fibers and reduce the influence of damage introduced from steps 1-3.
Resulting sample dimensions were 5.125 𝑖𝑛 and 1 𝑖𝑛 (nominal).
5. The samples were then notched using a 0.25 𝑖𝑛 3-flute “diamond-like” coated ball
end mill shown in Figure 15 below.
a. This was done for 5 samples at a time and cut as slowly as possible.
Typically, a machinist would take multiple passes to produce such a deep
cut. However, to prevent the introduction of damage during the notching
process, this was done in a single pass.
Figure 15. Diamondlike-Coated Carbide End Mill, Ball-End, 4 Flute, 1/4" Mill Diameter,
2-1/2" Overall Length [19]
53
After the samples were cut to the appropriate dimensions, they were cleaned and
dried using only water to remove oils and particulates from the surfaces. Prior to testing,
the samples still needed further preparation to use in the IPL and for the use of ARAMIS
DIC. Each sample was tabbed, not necessarily to reduce gripping effects but to ensure each
sample was thick enough (0.125 𝑖𝑛 minimum) so that the transverse support plates in
Figure 6 could be used for each test.
Table 4. Statistical values for geometry of all tested coupons. Measured with Mitutoyo
digital calipers verified with standardized machinist gauge blocks
Measured Property [spec (in)] Statistical Values (in)
Mean Std. Dev
Notch Depth [0.50] 0.5036 0.0050
Notch Width [0.25] 0.2497 0.0013
Coupon Width [1.00] 1.0032 0.0083
Samples were then painted using a flat white and flat black self-etching spray paint.
The white spray paint provided a base coat that was then “speckled” with the black paint.
This produced a stochastic pattern satisfactory for the DIC software to establish facets for
strain and displacement calculations. Self-etching paint was chosen as other paints would
chip off the surface of the coupon during deformation due to poor paint-to-coupon bonding.
An example end-result of coupon manufacturing and preparation can be seen in Figure 16
below.
54
Figure 16. Result of manufacturing procedure. *This sample was flawed due to improper
tabbing and was therefore not tested. A side-view to show tabs (top). Front view (middle).
Inch ruler for scale (bottom)
55
EXPERIMENTAL DATA – MULTIAXIAL TESTS
The primary purpose of this data is to produce qualitative and quantitative data to
inform and validate progressive damage models. Only some of the data produced from
these tests can be shown in a document format. Most of the data obtained from these
experimental tests is stored as ARAMIS files that may prove useful for direct comparison
(in ARAMIS) of progressive damage models and actual tests. Later software versions of
ARAMIS allow for the import and direct comparison of analytical data to measured
experimental data (these comparisons have not been performed for this research).
Experimental data is also exported as video report files like Figure 12 shown previously.
These video files may prove useful for qualitative comparisons between analytical models
and measured experimental data.
The data that will be shown focuses on damage initiation under multiaxial loads.
As discussed previously, composite materials’ constitutive responses are not well-
understood and their damage characteristics even less so. Providing information regarding
damage initiation may be a useful tool as the onset of damage is just as poorly understood
as its progression throughout the material.
Testing Procedure
After coupons are fully prepared, they are then ready to be tested. This section will,
in little detail, discuss the steps necessary for testing a sample using both the IPL and
ARAMIS DIC systems simultaneously. It is important to note, like all other previous
works with the MSU IPL, that the IPL has not been optimized for high volumes of tests
56
and therefore each test requires a minimum of thirty minutes to perform. The basic steps
for running a test in the IPL are listed below:
1. Load a fully prepared sample in the IPL grips
a. Similar to loading a sample in a typical Instron machine, the crosshead is
moved into position and the sample is clamped in place.
b. Clamping the sample is done incrementally. Some load is applied via the
hydraulic pump, fasteners tightened, and transverse support plates slid up
against the side of the coupon. Doing this incrementally ensures all gripping
components are in full contact with the sample.
2. Recalibrate IPL position and initialize test.
a. IPL position is recalibrated to ensure current configuration is accurately
being controlled by the software.
b. Initialize IPL test speed and begin load output voltages.
3. ARAMIS DIC project file is created and any necessary adjustments (shutter time,
lighting, etc.) are made.
4. Proper displacements and data storage location (for .txt file) are chosen on the IPL.
5. ARAMIS DIC is started and begins taking images (stages) and collecting load data.
Several stages are taken before starting movement of the IPL.
6. Once the ARAMIS DIC measurement has been confirmed, the IPL is started and
the test runs to the prescribed displacements in step 4.
7. Once final failure occurs, or for any reason the test needs to be stopped, ARAMIS
is stopped first then the IPL is stopped.
57
Loading Paths
The theory for testing these samples was similar to the original proposition by the
NRL [9]. Through many “arbitrary” combinations of in-plane displacements, a material’s
damage characteristics may be fully characterized for all displacements encompassed by
tested values. The mechanical behavior of the material is understood to depend only on
the current internal strain state and to be independent from any particular path. This path
independence allows for ‘a family of loading paths to be selected solely on the basis of
convenience’ [9].
NRL’s publication goes on to describe and justify their load path selection (Figure
17 below) and state that due to material symmetry ‘only the half-space corresponding to
positive sliding displacement [x-displacement] need be considered’. However, due to
limits of the IPL and the lack of material symmetry of these samples, a different loading
path scenario was created for this study.
Figure 17. Schematic of NRL’s loading path definitions [9]
58
First, compression tests (−𝑑𝑦) are difficult with the MSU IPL. This is due to
several reasons but is primarily due to out-of-plane response of the IPL, and buckling due
to coupon geometry. Buckling is said to be due to coupon geometry since buckling is
geometrically stability driven and not driven by material properties. Several compression
tests were carefully performed in the IPL to assess the IPL’s ability to perform compression
tests. Each compression test that was attempted failed due to an out-of-plane response of
the IPL and ultimately buckling of the coupon. Buckling of the coupon could be rectified,
in theory, by changing coupon geometry for compression tests and out-of-plane response
of the IPL may be rectified by adding stiffeners to the out-of-plane constrainers. However,
for this research, buckling tests were omitted as they did not produce reliable in-plane strain
data. An example of coupon buckling can be seen in Figure 18 below showing an out of
plane displacement of the coupon from 0.494 𝑚𝑚 to −1.081 𝑚𝑚, a 0.062 𝑖𝑛𝑐ℎ out-of-
plane displacement of the sample. Again, since the ARAMIS cameras are stereo cameras,
they allow for displacement calculations out-of-plane (z-direction) but cannot accurately
compute non-planar strains.
59
Figure 18. Image of out-of-plane displacements recorded via DIC for a failed compression
test
Second, as mentioned above, material symmetry does not exist for off-angle
laminates about the x-axis. Therefore, it is important to include both positive and negative
x-displacements for the load path selection. The resulting load path selection is similar to
NRL’s except for these changes and can be seen in below in Figure 19. The load paths
included in the figure are only those performed on every laminate. Some laminates have
more tests included for the data presented later. For more information on all tests
performed, refer to the attached test matrices located in the appendix. Note that
displacements recorded in the test matrices are not actual displacement achieved and were
60
entered as larger-than-necessary values to ensure final failure for each test. For more
information regarding tests performed in this study, see the test matrices in APPENDIX A.
Figure 19. Normalized load paths (displacement paths) performed for every laminate.
Vectors shown as unit vectors to display direction only
61
Data Acquisition
As mentioned before, the ARAMIS system acts as the primary data acquisition for
testing. All data collected is stored within ARAMIS software and can be handled in the
software as well as exported. Due to the version of the software (v6.3) automation and
calculation within ARAMIS is limited and can take more than two hours per sample to
fully process. See the recommendations section for more information regarding ARAMIS
software restrictions.
Since all the data is handled in ARAMIS but processing within the ARAMIS
software is cumbersome, some processing is performed in ARAMIS then exported and
passed on to MATLAB for further processing. This section will focus on the how the data
is obtained from the IPL to ARAMIS, post-processing within ARAMIS, and export
processes.
1. The stereo camera, as described previously is setup such that the entire coupon
is within view and will stay within view of both cameras for the duration of the
test. Each camera has been focused, and adjusted appropriately such that both
camera images are aimed close to the same point on the sample.
2. While the test is run, analog voltage values are collected via typical BNC cable
connections like an Instron machine. Three independent voltages are collected
with a range of ±10 𝑉. These voltages correlate to real-time load data from the
IPL and represent loads in each resolved direction (𝑥,𝑦, and θ). This coordinate
system is the same as discussed above and below (ex. Figure 3). These voltages
are then linearly transformed with independent gain values to convert the raw
62
voltages to units of 𝑙𝑏𝑓 for x and y direction loads and 𝑖𝑛 ∙ 𝑙𝑏𝑓 for the moment
about the top center of the coupon.
3. Once a test is completed, data collection in the ARAMIS software is terminated
and processing begins. There are several basic steps that must be performed to
each test before actual processing begins.
a. Several “start points” are created on each area of interest on the coupon.
A minimum of one start point must be created for each intact part of the
gauge section; once cracking occurs, a start point must exist on each
side of the crack to process both areas.
b. After all the desired start points are created, the project is then
computed. This step can take up to an hour for each test depending on
gauge section size, facet size, and length of the test. For these samples,
facets of 15 × 19 pixels were used and a test length of 150 stages was
the desired test length. This was to keep processing time to a minimum
but maintain reasonable data density.
4. The “compute” function described in 3b performs most of the necessary large
computations for the test. However, there are several additional steps within
ARAMIS software before the data can be exported for further post-processing
in MATLAB. Each of the listed items below can be seen in Figure 20 below.
63
Figure 20. Left image of test 11_024 as a demonstration of ARAMIS image acquisition.
(Green line [MC]) shows "movement correction". (Red line [VE 2]) shows “virtual
extensometer”. Coordinate system is shown at the notch tip
a. First, the test is corrected for “movement correction” which accounts
for rigid body displacements of the sample (grip slippage, light
distortions, IPL vibrations, and camera movements can be accounted
for). This step “fixes” the bottom boundary of the coupon by creating
an artificial fixed boundary condition on the coupon for computation
purposes.
64
b. Second, a coordinate transformation is then applied to the coupon. This
aligns the coordinate system about the center of the notch tip with the
y-axis pointing upward, the x-axis toward the open end of the notch, and
the xy-plane aligned with the plane of the sample. This ensures that
computed strains are consistent between each sample and correspond
directly to the finite element model discussed later.
c. Finally, a “virtual extensometer” is created at the top boundary of the
gauge section to directly measure the top boundary displacement of the
coupon. Again, the bottom boundary condition is accounted for in step
4a (movement correction).
5. All computed data in ARAMIS may be exported without further processing
within the ARAMIS software. For reports, strain images, and other ARAMIS
objects, further processing within ARAMIS is required.
Post-Processing in MATLAB
Basic Post-Processing Scheme
For completeness and reproducibility, this section will describe, in fair detail, the
post-processing scheme used to further analyze test data. Some of the post-processing is
handled directly in the ARAMIS software as described above. However, due to software
and automation restrictions, further post-processing of data is handled in MATLAB. The
basic post-processing scheme is described below and, for clarity, relative figures will be
included in-line with relevant text:
1. Two methods of exporting processed data from ARAMIS to .txt files
65
a. First, a single text file including displacement data from the ‘Virtual
Extensometer’ created in step 4b above and collected load data for all
stages obtained in step 2 above.
b. Second, individual text files for each stage include facet location
(deformed and undeformed), and facet strains (deformed).
2. Displacements and loads from the text file created in step 1a are imported to
MATLAB to determine dissipated energy.
a. Numerical integration for each load path (x-displacement, y-
displacement, and rotation) to determine dissipated energy developed
throughout the duration of the test. This is performed using a standard
trapezoidal rule numerical integration scheme shown in Equation (6) as
a demonstration of the integration scheme for the x-component.
𝐷𝐸𝑥 = ∫ 𝐹(𝑥)𝑑𝑥
𝑛
𝑜
≈∆𝑥
2[𝐹(𝑥0) + 2 ∑ 𝐹(𝑥𝑖)
𝑛−1
𝑖=1
+ 𝐹(𝑥𝑛)] (6)
b. Calculated dissipated energy determines the stage where damage is
initiated during the test. This calculation and result has already been
demonstrated in Figure 9.
3. Based on the damage initiation stage determined from 2b, strain data from 1b
(for the appropriate stage) is imported and further processed in MATLAB.
a. Strain data is truncated to exclude “exterior” facet calculations. Facets
close to the edge of the sample are typically inaccurate as “boundaries”
are not well-defined for these edge facets. All facets lying outside the
66
inner 90% for each left-to-right and top-to-bottom measurement is
discarded. See Figure 21 below for a graphical representation of the
acceptable facet location.
Figure 21. Image of facet truncation. Area enclosed in the green rectangle is the area of
accepted facets
b. All truncated strain data for each sample type (each individual laminate)
is collected and grouped as a whole.
4. A point cloud is created in 3D strain space for each laminate (all components of
in-plane strains). Resultant figures will be displayed in the Results section to
follow.
5. Point cloud density is calculated. This is done for every single point in the point
cloud – each point is assigned a “weight” based on the point cloud density at
67
that point. This calculation takes approximately an hour for each laminate but
only needs to be performed once per point cloud.
a. The distance is calculated from a single point to every other point in the
point cloud. This is done with the simple distance formula shown below
in Equation (7) below. As stated, this calculation is performed for every
point therefore requires 𝑛×𝑛 calculations where 𝑛 is the number of
points in the point cloud.
𝑑𝑖 = √𝑥𝑖2 + 𝑦𝑖
2 + 𝑧𝑖2 (7)
b. For a single point, after each distance is computed, these distances are
sorted and compiled into a single value using weights seen in Equation
(8) below. The weights applied can be controlled arbitrarily to produce
varying results for the next section. The weights displayed below are
the values used for the figures produced and shown in this report. Note,
these 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 values are actually distances; therefore, the smaller the
𝑑𝑒𝑛𝑠𝑖𝑡𝑦 value, the more dense the point cloud at that point.
𝑑𝑒𝑛𝑠𝑖𝑡𝑦 = 𝐶1 × 𝛿1 + 𝐶2 × 𝛿2 + 𝐶3 × 𝛿3
𝛿1 = (∑ 𝑑𝑖
50
𝑖=1
) 50⁄
𝛿2 = ( ∑ 𝑑𝑖
0.005×𝑛
𝑖=1
) (0.005×𝑛)⁄
𝛿3 = (∑ 𝑑𝑖
𝑛
𝑖=1
) 𝑛⁄
(8)
68
i. The average of the smallest 50 distances to the point (𝛿1).
ii. The average of the smallest 0.5% of the distances to the point
(𝛿2). This results in approximately the smallest 2500 distances.
iii. The average of all distances (𝛿3)
iv. For these results: 𝐶1 = 2, 𝐶2 = 2, 𝐶3 = 1. This is to ensure
closer points have a greater weight than all points.
c. Finally, this process repeats for every single point in the cloud and
assigns a 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 value to each. This is where the calculation takes
most of its time and requires 𝑛×𝑛 calculations.
6. Using point location (휀𝑥, 휀𝑦, 휀𝑥𝑦) and the 𝑑𝑒𝑛𝑠𝑖𝑡𝑦 computed in step 5,
isosurfaces are created based on the cumulative distribution probability of each
point. Using this cumulative distribution probability, the calculated 𝑑𝑒𝑛𝑠𝑖𝑡𝑦
value has no weight and has no relative significance compared to any other
value. The surface can now be controlled entirely based on point density
probability, i.e. the surface lies on or contains the densest XX% of all data
points.
7. Isosurfaces can then be “sliced” and projected onto planes for easier viewing.
This concludes the primary post-processing performed for this study. Each step
was performed on every laminate type tested; selected results will be displayed in the
results section. The schematic version of the above post-processing scheme is displayed
in Figure 22 below.
69
Figure 22. Schematic view of the post-processing scheme used to further process ARAMIS
data via MATLAB. For more detail about each step, refer to the section above
70
The strain data for all laminates of a single material can then be computed and shown; this
is a step in the post-processing that is not displayed in the schematic above. The global
coordinate strain data can then be transformed into each ply and then displayed as a “ply
property” with axes 1, 2, 12 in-plane strains. This may prove to be more useful from a
modeling perspective but, requires manipulation of the data and assumptions to resolve.
• First, the assumption used to produce this ply-level data is that strains are
constant through the thickness of the sample. This is not necessarily true and
may produce unrealistic results.
• Second, and simply, this requires a simple coordinate transformation of the
measured surface strain data onto each other laminate. For example, the quasi-
isotropic laminate would need 4 strain transformations to account for each layer
because the layup is [−45/90/45/0]𝑠. This transformation can be seen by the
simple linear transformation of the strain vectors shown in Equation (9). 𝜃
represents the angle of rotation from the global x-axis to the fiber direction of
the respective ply.
{
휀1
휀2
휀12
} = [𝑐2 𝑠2 −2𝑐𝑠𝑠2 𝑐2 2𝑐𝑠𝑐𝑠 −𝑐𝑠 𝑐2 − 𝑠2
] {
휀𝑥
휀𝑦
휀𝑥𝑦
}
Where 𝑐 = 𝑐𝑜𝑠(𝜃) & 𝑠 = 𝑠𝑖𝑛(𝜃)
(9)
After each ply-level strain has been determined for every ply within every laminate, all ply
data is gathered together as either IM7/8552 or Toray ply-level strains (whichever the ply
belongs to). This data is now much larger than any one laminate and cannot be processed
the exact same way; but it is processed similarly.
71
Results and Discussion
Again, results from experimental tests are to produce data that informs and validates
progressive damage models. As such, this data is qualitative and quantitative. Examples
of the qualitative data can be seen below that may be useful for side-by-side comparison to
finite element models. Results like those produced by the post-processing scheme above,
are quantitative and could be useful for informing progressive damage models in different
ways. Progressive damage models including discrete damage models** but, particularly
continuum damage models†† require a failure criterion to determine when the element
experiences damage. As discussed previously, existing analytical failure criterion are not
necessarily representative of how the material experiences damage; the failure criterion
developed in this study and described below is entirely based on experimental data.
Failed Coupons
First, a lot of information can be obtained by inspecting failed coupons. Fracture
surfaces are often observed in metal materials to better understand how fracture occurred;
the same can be done for composites. Images of some of the most interesting failed
coupons can be seen in Figure 23, Figure 24, and Figure 25 below. Failure modes are
typically consistent between laminates and do not drastically vary between ply materials.
For this reason, two images Figure 23 and Figure 24 are for the IM7/8552 material and
** Discrete Damage Models – Models that attempt to discretely model damage in the material. Done by
“breaking” elements. These models can be particularly accurate on a small scale as they may model
damage directly but require the most computational expense to solve. †† Continuum Damage Models – Damage prediction models that model damage by a means of “degrading”
material properties of damage elements. In practice, these models are the most efficient from a
computational standpoint but the least accurate in material representation especially on a small-scale.
72
Figure 25 for the Toray material system. These figures do encompass all layup orientations
included in this study.
Figure 23. Laminate 11 interesting failures. Written labels correspond directly to the test
matrix. For scale, coupon widths are 1 inch or refer to Figure 14 above
Figure 23, above, displays interesting failures of the IM7/8552 quasi-isotropic
laminate [-45/90/45/0]s. Since this laminate is composed of 4 different layers, many
damage modes can be seen. The 3 selected coupons demonstrate, in different ways, some
of these failures. The top coupon shows the most failure modes including fiber rupture
seen in the 0° plies, matrix cracking in every ply, and delamination. The middle coupon is
interesting as there is clear fiber rupture in the −45° plies. The bottom coupon, although
73
it is difficult to see, has not failed entirely yet and is held together by the intact 0° plies.
Again, this image provides interesting data into many failure types.
Figure 24. Laminate 13 interesting failures. Written labels correspond directly to the test
matrix. For scale, coupon widths are 1 inch or refer to Figure 14 above
Figure 24, above, displays interesting failures of the IM7/8552 specially orthotropic
laminate [0/90/0/90]s. This laminate is composed of only two types of layers, 0° plies and
90° plies. Due to this, the failure modes are limited. Matrix cracking usually occurs in the
0° plies at the notch tip; this is expected as there are large shear stresses occurring on these
plies, in this location during any kind of tension. Each coupon shoes failures in different
ways. Fiber rupture and matrix cracking between fibers are evident on every sample. As
These samples typically experience fiber ruptures in large clusters of fibers at the same
74
location. This is likely due to transverse cracking in the 90° plies progressing into the 0°
plies and causing a weakened area. When fibers rupture, stored energy must be distributed
and often causes catastrophic failure of the surrounding area.
Figure 25. Laminate 4 interesting failures. Written labels correspond directly to the test
matrix. For scale, coupon widths are 1 inch or refer to Figure 14 above
Figure 25 displays several of the interesting failures experienced by the Toray
material system’s [-45/45/-45/45]s laminate. The top coupon displays a case where the
−45° plies experienced fiber rupture. The middle coupon displays large amounts of matrix
cracking along the fibers and delamination. Finally, the bottom coupon displays an
interesting kink-band looking failure of the −45° plies. Typically kink-band formation is
associated with compression; this coupon did, in fact, experience compression due to the
75
rotational component of the load path. From these images, some information regarding
damage propagation and especially final failure can be determined. Furthermore, closer
examination such as optical microscopy may be performed to provide more insight.
Digital Image Correlation Results
These results depict only several of many images that may help to inform and, in
particular, validate progressive damage models. These images are only still frames from
videos which cannot be displayed in this document format. These videos provide data such
as test time, loads, displacements, and strains throughout the duration of the test. This
duration is the entire test from the start, through damage initiation and propagation, and
including final failure if the sample experienced final failure. Even a single image from
every sample would prove to be too much to display in this document format and, for this
reason, a selected few images from several samples that show interesting damage are
displayed.
First, as an example of a single test, there will be three images shown from a single
test. These images will include the stage of damage initiation (Figure 26), an intermittent
stage of depicting progressed damage (Figure 27), and a stage right before final failure
showing damage in its most developed state (Figure 28). The sample that was chosen to
demonstrate this was an IM7/8552 quasi-isotropic laminate [-45/90/45/0]s (as tested).
Each image will be a report image of the “Major Strain” or maximum principal strain.
76
Figure 26. Test 11_021 image of Major Strain report at damage initiation
From Figure 26 above, the major strain clearly shows a higher strain development
near the notch tip. From this figure alone, a side-by-side comparison to a model may
validate the model if the model produces similar principal strains in the same locations
created by the same loads. Not only would this validate the damage initiation of the
material but also the constitutive response and the properties input to the model.
Furthermore, additional information may be conveyed using the same report format such
as 휀𝑥, 휀𝑦, or 휀𝑥𝑦. Along with this, nearly any information can be conveyed in this format
from ARAMIS as long as it is measured by ARAMIS and is not a calculation.
77
Figure 27. Test 11_021 image of Major Strain report at an arbitrary intermediate damaged
state
Obviously, this is an arbitrary intermittent damaged state of the coupon somewhere
between damage initiation and final failure. It is clear from the figure that cracking is
starting to form on the outside −45° ply and this will become more evident in the next
images. Also from this image, where the outside crack is progressing, another high-strain
region depicts a crack forming – this is likely due to cracking forming in the 3rd layer in
(+45° ply). It may also be assumed that transverse cracking is occurring in the 90° plies
at the same location and that potential delamination has begun. Of course, these images
may fail to provide exact failure mode but for the validation and comparison to a finite
element model, these may prove satisfactory.
78
Figure 28. Test 11_021 image of Major Strain report at the stage before final failure
Thirdly, and finally for this example, the coupon has failed. The specimens
typically lose paint from the outside surface due to the damage of the coupon itself. This
makes for difficult strain measurements on and near this region. However, these images
may not be entirely useless as the final failure mode may be evident from this image and
much of the data collected to this point provides information regarding when, where, and
how damage is working its way through the material. Progressive damage tables such as
that shown in Table 5 below are developed from the above ARAMIS images. Complete
tables for each laminate and every test are displayed in APPENDIX C.
79
Table 5. Snippet of progressive damage table as developed from the images above. For the
full table and for each laminate, refer to APPENDIX C
Each damage mode labeled in Table 5 above is determined solely based on
ARAMIS calculations such as major strain (maximum principal strain) direction and the
observation of failed samples. The data was then interpreted analytically using the finite
element model, which will be described further later. All damage modes included are only
damage modes that could be directly observed from the data. For this reason, some listed
failure modes were broad and not specifically defined. Abbreviations for the interpretation
of the table are as follows:
• MC – Matrix Cracking
o Cracks begin in the matrix, propagates, and leaves fibers intact.
Typically observed as the first damage. High strains and maximum
principal strain directions provide information regarding this type of
damage. Further details into this mode of failure (such as shear or
tension) are not easily determined and therefore not included in the
resultant tables.
• DL – Delamination
stage ply mode stage ply mode stage ply mode
11_003 101 -45 MC - LNT 128 All DL 143 0 FR - LNT
11_004 72 90 MC - LNT 84 45 MC - LNT 125 0 FR - LNT
11_005 90 -45 MC - LNT 95 90 MC 107 All DL
11_006 42 -45 MC - LNT 61 90 MC 153 0 FR - LNT
11_007 58 0 MC 85 -45 MC - LNT 112 All DL
11_008 62 90 MC - LNT 86 OA MC - LNT 105 All DL
11_009 58 45 MC 148 -45 DL 283 All DL
Damage initiation Intermittent Stage Final FailureTest
Name
80
o Often interpreted from out-of-plane displacements calculated by the
DIC. This failure typically causes final failure for tests with large
rotation components as fiber failure is observed for these
displacements less often.
• FR – Fiber Rupture
o Fibers incur damage directly causing fiber breakage. Typically, the
“last straw” as stored energy suddenly released into the sample
causes catastrophic failure.
• OA – Off Angle Plies
o All non-zero and non-90 degree plies.
• LNT – Localized at Notch Tip
o Damage location is confined primarily to the notch tip. If not
specified, damage occurs multiple places.
• TG|BG – Top Grip or Bottom Grip
o Damage is occurring locally at the top grip or bottom grip.
It is important to note, that although digital image correlation may measure and
return computed strains, that strain measurements may be inaccurate for representing the
entire sample especially through its thickness.
81
Failure Surfaces
The primary quantitative results of this study are shown below. Again, like the
qualitative results, all quantitative results can simply not be shown as there is simply far
too much data. As a note, in-plane shear strain values are displayed in figures as
percentages. This is unconventional but was performed for visualization purposes and does
not have a physical meaning like the normal strains in the x and y directions.
Importantly, the results will undergo a lengthy discussion describing their origins
and how they can be interpreted. As seen in the post-processing section, these results are
generally displayed as a probabilistic view of strain data. This strain data, again, is taken
from an experimental test result at the determined damage initiation stage. This stage is
where mechanical damage (material softening) is determined to have started during the
test. The DIC processing creates facets over the face of the gauge section and can be treated
like elements in a finite element model where each facet has a determined strain state
(휀𝑥, 휀𝑦, 휀𝑥𝑦). This multiaxial strain state (in-plane strain state) makes up the axes of the
figures depicted. An example of an individual facet’s strain state is shown below in Figure
29. This data is shown for all facets of all tests of a certain laminate at damage initiation.
As a note, before displaying these results, the displayed failure surfaces are not
geometry dependent and may be treated as material ‘damage’ properties. Even though the
coupon geometry was consistent in the creation of these surfaces, the measured data
(strains) do not depend on geometry. This is useful as the determined failure surfaces may
be useful in the validation or qualification of complex structures that utilize the appropriate
laminate-level and perhaps ply-level materials.
82
Figure 29. Image depicting a selected facet from an arbitrary test as seen in ARAMIS.
"Stage Point Data" displays all facet properties
It has been briefly displayed in the post-processing scheme that a point cloud is
created with all of the data for every test of a particular laminate. The stage used from the
test data is the stage of damage initiation as determined by dissipated energy calculations.
Dissipated energy calculations can detect only mechanical damage and cannot detect
physical damage. For this reason, the failure surfaces computed from this point cloud can
be treated like the other failure criterion mentioned previously that predict the onset of
mechanical damage/ material softening. However, since these calculations only account
for mechanical change, it is important to note that other methods must be used if an
understanding of physical damage is desired.
83
Figure 30. All in-plane strain data points collected for MSU - 11 tests. This includes all
tests displayed in the test matrices
Figure 30 above is a demonstration of what the point-cloud looks like. This was
shown briefly in the post-processing scheme but is more clear by itself. There are three
axes defined by the in-plane strain (휀𝑥, 휀𝑦, 휀𝑥𝑦) of each individual facet. After going
through much of the post-processing scheme, these data points can be shown based on their
where they lie in the cumulative probability distribution. From this, isosurfaces can be
created and shown based on these assigned values (shown as colors above). Notice the
color bar scale ranges from 98% to 100% and the inner-most points are packed so tightly
that approximately 98% of the data is not worth showing relative densities. The isosurfaces
created from this figure are shown below in Figure 31.
84
Figure 31. MSU - 11 nested isosurfaces defined by data density. Scatter points are also
shown for reference
From Figure 31, it can be further seen that data density increases drastically. Three
separate isosurfaces have been created and displayed; as an example, the red isosurfaces
labeled ‘99.3% densest data’ in the legend means that 99.3% of all MSU-11 strain data
lies on or within this enclosed surface. This is useful to create “failure surfaces” analogous
to failure criterion but, again, are based on experimental data. All failure surfaces displayed
are defined by the onset of damage, not final failure. Since these are truly measured data
from experimental tests, the material has experienced these strains without accumulating
85
mechanical damage‡‡. Probabilities are used to define these nested isosurfaces to isolate
outliers caused by error in digital image correlation measurements as well as create
different “layers” that offer different levels of conservativeness.
These nested isosurfaces can then be displayed based on how they look on each
primary plane. For the sake of brevity, only Figure 32 below will be shown in detail and
the remaining “slices” will be displayed at once in Figure 33. This is only an example for
the IM7/8552 quasi-isotropic laminate (MSU – 11). These “sliced” views are similar to
existing failure criteria but do not have any reliance on analytical assumptions. The three
nested “levels” are shown to illustrate the different levels of conservative criteria.
Since these surfaces are analogous to failure criteria, different levels of
conservativeness will be described in terms of failure criteria. As shown previously,
interacting criteria are typically more conservative than maximum stress and maximum
strain criteria in areas where combinations of stresses occur. This is due to the “rounded”
shape of the interacting criteria. Like these criteria, the different levels of isosurfaces are
displayed for the sake of conservativeness/ conservation. The blue inner surface displays
the most conservative surface based on the densest data and the smallest failure volume.
The red outer surface displays the least conservative surface as it allows less dense data to
define a larger failure volume. The smaller the volume in strain space, as displayed, the
more conservative the failure surface. This will be discussed further in the conclusions
portion to follow.
‡‡ Mechanical Damage – There are generally two types of damage: physical and mechanical. Physical
damage is typically damage like micro-cracking that does not affect mechanical properties of the material.
As physical damage develops further, larger cracks are formed that affect mechanical properties of the
material – this is mechanical damage.
86
Figure 32. Example of nested isosurfaces displayed on the eps x = 0 primary plane
To reiterate, as described in the post-processing scheme, these surfaces are
generated based on measured strain states of mechanically undamaged coupons (where no
mechanical damage has occurred). The data used to define these surfaces is taken at the
last undamaged state of the material during the test. In theory, disregarding outliers of
digital image correlation error, the material can experience these strains without incurring
mechanical damage.
87
Figure 33. Nested isosurfaces for MSU - 11 displayed on each primary plane. Original
nested isosurfaces (top-left). Primary plane eps xy = 0 (top-right). Primary plane eps y =
0 (bottom-left). Primary plane eps x = 0 (bottom-right)
As shown, these surfaces are material failure predictive surfaces with different
levels of conservation. The results above only pertain to the laminate as a whole and should
not be considered representative of any laminate other than the IM7/8552 quasi-isotropic
laminate it was derived from. Also, these surfaces should not be considered representative
of individual ply properties on the described 1, 2 and 12 planes. These figures have been
generated for every laminate in this study.
88
Figure 34. MSU - 1 nested isosurfaces defined by data density. Scatter points are also
shown for reference
To aid in demonstrating the differences between the two material systems, Figure
34 above displays the result for the Toray material system’s quasi-isotropic laminate. This
figure was created with nearly as many tests as the IM7/8552 material system as well as
the same test load paths. This figure is to be compared to Figure 31 above (the IM7/8552
version of the laminate). As shown in Table 3, these laminates share the same stacking
sequence but are composed of different ply materials.
89
Figure 35. Nested isosurfaces for MSU - 1 displayed on each primary plane. Original
nested isosurfaces (top-left). Primary plane 휀𝑥𝑦 = 0 (top-right). Primary plane 휀𝑦 = 0
(bottom-left). Primary plane 휀𝑥 = 0 (bottom-right)
Furthermore, these isosurfaces can be compared with each other side-by-side based
on how the surfaces intersect the primary planes. Figure 35 above shows the Toray material
system’s quasi-isotropic laminate isosurfaces “sliced” and viewed on each plane. This
figure is to be compared to Figure 33 above which, again, where the same image for the
90
IM7/8552 material system is shown. Furthermore, Figure 36 below shows this comparison
further. The difference between the two material systems is evident by the isosurfaces
displayed. It is understood that the Toray material system is more brittle than the IM7/8552
material system and this is evident from comparing the experimental failure surfaces
created. The Toray material system results in a smaller overall volume and the isosurfaces
do not extend as far as the IM7/8552 system’s. This demonstrates that the Toray material
system, in general for this laminate, does not reach as high of strains before the material
experiences failure.
Figure 36. Direct comparison of MSU - 11 tests (Figure 33) shown in blue and MSU - 1
tests (Figure 35) shown in red
As discussed at the end of the post-processing scheme section, data was then
transformed into local ply coordinates. This produces strain results that pertain to each
material system ply-level. Before displaying results, it is important to reiterate that the
following results may have errors due to the assumption discussed in the post-processing
section. Figure 37 and Figure 38, below, show the ply-level results for the Hexcel
IM7/8552 material system and the Toray material system respectively.
91
As a note, the results obviously differ from the results provided above for the
laminate-level failure surfaces. This is mostly due to the size of the data after it was
transformed and collected for each material. The resultant data set was so large that it
would take approximately 70 hours to process each. So, as a result, the “raw” point cloud
is shown without weights in the top left of each figure which doesn’t provide much insight
about the material without, again, taking these “slices”. The “slices” are also displayed in
each figure with a slightly different format than previously shown but equally as useful.
92
Figure 37. Ply-level "failure surfaces" produced from original point cloud of the Hexcel
IM7/8552 material system transformed and resolved into each ply's local coordinate
system. Original point cloud (top-left). Primary plane 휀12 = 0 (top-right). Primary plane
휀1 = 0 (bottom-left). Primary plane 휀2 = 0 (bottom-right)
93
Figure 38. Ply-level "failure surfaces" produced from original point cloud of the Toray
material system transformed and resolved into each ply's local coordinate system
94
FINITE ELEMENT MODEL
Model Definition
This section will focus on the definition of the finite element model developed using
ANSYS Workbench. It is important to note that although the data from this research is to
inform and validate progressive damage models, the following model is not a progressive
damage model. The model to be described is a linear elastic model created only to aid in
experimental data interpretation. This model is important for two reasons. First, if a linear
elastic model can produce results resembling a test up to damage initiation then the next
step can be taken to produce a model that can compute damage. Without first confirming
a simpler model, attempting to model more complex behavior would be moot. Second,
and to be discussed further later, the following model will act as a supplement to
experimental data where the ARAMIS DIC falls short.
As discussed briefly in the Experimental Data section, edge facets computed by
DIC are often inaccurate or impossible to calculate. This is due to the lack of boundary
definitions for these edge facet (again, similar to elements). Without accurate references
to define these “element” boundary conditions, the displacements, and therefore strains,
are impossible to accurately compute. So, as stated above, this finite element model
supplements digital image correlation data by allowing a representative model accurately
compute strains of elements close to edges. Data processed further will include all strains
at node locations which allows for strain “measurements” on the edges of the model. Also,
for this reason, the model has been created with smaller elements around the notch tip to
provide higher-resolution information near this area of interest.
95
Model and Layup Setup
ANSYS Workbench is actually just a graphical user interface to ANSYS APDL that
provides extra tools and “modules” to be applied to the model. Therefore, logically the
workflow of Workbench operates the same way as APDL. This can be shown graphically
from the top-level view of Workbench shown in Figure 39 below. Each module included
in the model design tree serves a different purpose. ANSYS Workbench was chosen as it
provides all features available in APDL as well as tools not included directly in APDL.
Workbench also offers robust automation ability which is useful for replicating many tests
using identical input and output formats. Again, for the sake of completeness, each module
will be discussed in detail below.
Figure 39. Top-level view of ANSYS Workbench model
Module A. Geometry
Module A. Geometry. The geometry construction of the model is made to mimic
geometry of experimental coupons minus the grip areas. Therefore, only the gauge section
was modeled as this was the area of interest. The model geometry and mesh result can be
seen in Figure 40 below. The mesh is refined around the notch to provide higher resolution
96
data as the notch is the primary area of interest. The coordinate system displayed matches
the coordinate system applied to each experimental sample within ARAMIS software; this
is to maintain continuity between the model and experimental data.
Figure 40. Model mesh definition with coordinate system shown
Module B. Engineering Data
Module B. Engineering Data. The engineering data module supplies material
property data to the model. Here is where the properties displayed in Table 1 and Table 2
(shown previously on page 48 and 49 respectively) are applied to the model. The next
97
module uses the material properties of individual plies to create each ply of the coupon
separately. Unlike some linear elastic mechanical models of composites, this method does
not require so-called “smeared properties” to be applied and is more robust as each ply is
handled separately.
Module C. ACP (Prep)
Module C. ACP (Prep). As mentioned above, each ply (layer) is handled separately
within ACP (Prep). This allows for the creation of shell elements representing individual
plies rather than solid elements with smeared properties that may be less accurate. Figure
41 below shows the distinction of individual ply elements applied to the geometry. Most
of the basic functions ACP (Prep) offers were used in the creation of this model including
the ability to define ply orientation, material thickness, coupling properties, strength
properties, and element set definitions.
98
Figure 41. ACP (Prep) figure showing ply layer definition, coordinate system, and
distinction of elements through the model thickness
Although not used in this model, ACP (Prep) also allows the creation of composites
comprised of many materials in any configuration, e.g. foam-core box beams or wind
turbine blades. Another advantageous feature is the definition of ply-drops and draping
coefficients for complex structures.
Module D. Static Structural
Module D. Static Structural. This is the main module that handles boundary
condition setup, solution and result information. This module gathers all inputs from
previous modules, inputs them into APDL, continues APDL setup, and finally solves the
99
APDL code. At any point in the design tree of ‘Static Structural’, manually edited APDL
code such as element definitions and export scripts can be added.
To solve a file in ‘Static Structural’, boundary conditions must be applied, and
desired solution information must be defined. The boundary condition setup is identical to
the schematic shown previously in Figure 3. Boundary conditions applied to the model
and the assumptions made are explained fully below.
Solution and results information can be displayed on a nodal or elemental basis for
nearly any desired result. Solution result sets for this model are handled outside of ANSYS
and are exported as text file. The process for the use of solution information from the
model is described further below.
Module E. ACP (Post)
Module E. ACP (Post). Appropriately named, ACP (Post) is the post-processor side
of ACP. ACP (Post) imports solution information from the ‘Static Structural’ module and
makes further calculations. This module does not model damage in any way but may be a
useful tool for preliminary failure calculations. Elements are analyzed individually in ACP
(Post) and user-defined or built-in failure criteria can be applied to each element. For
example, Maximum Strain failure criterion can be applied to the model and each element’s
“failure prediction” based on the applied criterion can be displayed. The image shown in
Figure 42 is an example of Maximum Strain applied to an arbitrary simulation. This
module was used in the aid of producing data displayed in the APPENDIX B.
100
Figure 42. ACP (Post) example of failure display using Maximum Strain criterion of an
arbitrary simulation. Image only displays predicted failure for selected layer (surface layer)
Parameter Set
Parameter Set. The parameter set module provides a means of accessing and
changing numerical or script variables within ‘Static Structural’ without the need to access
the full ‘Static Structural’ module. This feature was used for automation purposes and will
be explained more below.
Boundary Conditions
As mentioned above, boundary conditions are applied in the ‘Static Structural’
Module. The bottom boundary is defined as a ‘fixed-rigid’ boundary condition. The top
boundary is defined as rigid but allows for both x and y component displacements and
rotation “about z”. The rotation component is applied to the centroid of the boundary and
therefore rotation is defined to rotate in the x-y plane about the top-center of the coupon.
101
Rigid behavior for each boundary condition simply restricts deformation of the boundary
it is applied to but allows rigid body displacement. Figure 43 shows an example of “pure”
mode displacements applied to the top of the coupon.
Figure 43. Example of Displacements applied to top boundary condition in ANSYS finite
element model. Undeformed (top left), Positive x-displacement (top right), Positive y-
displacement (bottom left), Positive rotation (bottom right)
Model Assumptions
There are several assumptions made in the model that must be addressed.
Assumptions are made to simplify complex systems to a point that they can be understood;
but, as a result, these simplifications are often not realistic and may cause errors in
calculations and ultimately errors in understanding. In a way, finite element models are by
nature an assumption and should be treated as such. To avoid irrelevant and improper
conclusions from this work, the primary assumptions of the finite element model are listed
and briefly discussed below.
102
• Linear Elasticity – Discussed briefly above, the model used is a linear elastic
model and may not “well” represent the material. It is generally accepted
that CFRPs are not linear elastic in nature especially after damage initiation
[9]. However, there is evidence to support that there is a non-linear elastic
response prior to damage initiation [3]. Without “exact” material
constitutive properties, non-linearity included, models may stray from good
representation of their experimental counterpart.
• Constant Strain through the Thickness – This assumption is also made for
the experimental data up to but not after damage initiation. It is assumed
strains on the surface of the sample are representative of the strains through
the thickness (each ply experiences the same strains). This is applied in the
model by tying nodes together. This assumption was made since it
simplifies, reduces degrees of freedom within the model, and ultimately
reduces computation time required to solve the model.
• Rigid Boundary Conditions – This is assumed for the model and is a
reasonably realistic assumption. Schmitt [14] explored whether or not rigid
boundary conditions were appropriate to apply and concluded that there was
no significant change in the solution between the two methods. Physically,
if this assumption were to be applied to an experimental test, it would mean
that the sample is perfectly gripped.
103
Processing
This section will briefly describe the processing procedure of the finite element
modeling data. This post-processing scheme is nearly identical to the procedure described
in the experimental data portion of this report and will therefore be summarized with
primary differences elaborated. However, making the model resemble a real-life test
requires much setup aside from the model-definition setup. This is described briefly below:
1. Data is exported from ARAMIS into usable files. This is the same as step 1b
from the experimental post-processing scheme. Files are stored for each
individual stage and include undeformed and deformed facet displacement data
as well as strain data.
2. From these files, created in step 1, boundary conditions are directly measured
from the sample’s face via ARAMIS calculations. This step ensures that
boundary conditions applied to the model are, in fact, boundary conditions
measured directly from the sample during the test. It is only necessary to
reproduce the test as soon as damage initiates and this process will use this
stage.
a. Results data is imported to MATLAB and is filtered for the top boundary
condition and bottom boundary condition facets based on facet location
in the undeformed configuration. An example of these facets can be
seen in Figure 44. This step accounts for any coupon-slippage during
the test and ensures the model boundary conditions are as close to exact
measured displacements as possible.
104
Figure 44. Example of the facets chosen for step 2 to determine boundary conditions for
the ANSYS model
b. After the facets have been isolated in the undeformed configuration, a
line is fit to them and their undeformed location is noted.
c. These facets’ locations are then found (via the .txt file) and a line is fit
to them with their deformed location noted.
d. MATLAB then calculates displacements necessary to reproduce the test
in ANSYS (𝑑𝑥, 𝑑𝑦 and 𝑑𝜃).
3. After the boundary conditions are determined, a properly formatted Python
script is created via MATLAB to automatically run ANSYS Workbench to
reproduce tests based on the boundary conditions determined in step 2d.
4. Desired strain values are stored in .txt file from ANSYS for all simulated tests.
The strain values obtained from the simulation (step 4) are treated identically to the
values from the experimental portion of this study. Even though these values obtained
105
from the simulation are not measured values from a test, they do provide more insight about
what happens to the sample close to the notch tip. Again, this is due the fact that DIC lacks
the ability to accurately measure boundary displacements of facets created on the edge of
the sample. There are large discrepancies between the model and the experimental data
that will be discussed further in the next section.
Results and Discussion
The finite element model does not provide as much “weight” in supporting
progressive damage models as experimental results. The results shown below are not to
be treated on the same level as the experimental data previously displayed since the
analytical model has been built based on the assumptions discussed above. Nonetheless,
the finite element model for this study has possibly produced relevant data to supplement
the experimental data. These results can be seen in several of the figures displayed below.
Just like an actual test performed with DIC, full-field strain calculations can be
demonstrated as a “temperature plot” on the surface of the elements. This is useful as a
model can be compared side-by-side to the actual data. The ability, from a qualitative
standpoint, of the model to reproduce tests accurately is shown in Figure 45 below.
106
Figure 45. An arbitrarily chosen sample (11_021) results compared between experimental
results and the finite element model. Maximum principal strain displayed
The figure displays the maximum principal strain of a sample. Boundary conditions were
double checked; displacements were identical. As explained by the processing scheme, the
model was controlled based on measured displacements from the experimental test.
Boundary load responses were also checked between both results with an error of 12% for
the load in the x-direction where the load was small and an error of 8.7% for the load in
the y-direction which dominates the loading of this sample. The results are displayed with
similar color scales for direct comparison and show reasonably similar strain distributions.
The highest strained areas are identical with matching values. However, there does seem
to be a discrepancy as the experimental data may include higher strains due to what appears
to be transverse cracking underneath the surface of the sample. This cracking was
determined, via dissipated energy (as discussed above), to not affect the mechanical
response of the sample at the stage the image was taken.
107
Also, for comparison of experimental data to the finite element model, Figure 46
below shows the same sample (at the same stage) but with shear elastic strain displayed.
This is to compare another strain measurement for the same sample. Again, the shear
prediction from the model qualitatively matches the measured data of the test. The strain
distributions, again, match and show the high-strain regions in the same locations as well
as very similar values. These qualitative results are useful for the interpretation of
experimental data as digital image correlation only provides information about the
measured surface and nothing underneath.
Figure 46. The same sample shown in Figure 45 (11_021) results compared between
experimental results and the finite element model. Shear elastic strain displayed
108
Figure 47. Point cloud developed from modeling all MSU - 11 tests and extracting strain
data. Displayed as before
Furthermore, point clouds may be created the same way as the experimental data
provided. The color formatting of the points also remains the same and is based on the
probability distribution of calculated point densities. An example of a point cloud is shown
in Figure 47 which represents the MSU-11 tests reproduced in the finite element model.
Even though the model reproduces tests well, as shown in Figure 45 and Figure 46 above,
there are obvious discrepancies between the experimental point cloud and the finite
element model point cloud.
109
Figure 48. Nested isosurfaces for FEM - 11 displayed on each primary plane. Original
nested isosurfaces (top-left). Primary plane 휀𝑥𝑦 = 0 (top-right). Primary plane 휀𝑦 = 0
(bottom-left). Primary plane 휀𝑥 = 0 (bottom-right)
Obviously, due to assumptions of the model (listed above in the “Model Definition”
section) and error in experimental data (primarily digital image correlation measurement
errors as discussed previously), the point cloud created by a model will not match the point
cloud generated by experimental data shown in Figure 30 and Figure 32 previously. An
110
explanation as to why the point clouds appear so different is that the model is truly an
idealized (“perfect”) version of the test. Figure 48, above, shows the nested isosurfaces and
their respective “slices” as similarly demonstrated with the experimental data. It is easily
seen from this image that the point cloud truly is concave primarily where (휀𝑥 < 0 & 휀𝑦 <
0 ) as well as the region where (휀𝑥 > 0 & 휀𝑦 > 0 ). This makes logical and physical sense
as the model does not allow for tension-tension and compression-compression of an
element due to the idealized Poisson effect§§.
Figure 49. Direct comparison of "sliced" isosurfaces of MSU - 11 tests (Figure 33Figure
35) in blue and the FEM of MSU - 11 (Figure 48) in red
The experimental point cloud data, on the other hand, obviously does not have the
“empty” regions where (휀𝑥 < 0 & 휀𝑦 < 0 ) and (휀𝑥 > 0 & 휀𝑦 > 0 ). This means that
calculated facets in experimental tests did, in fact, experience strains of dilatational
expansion and contraction. This discrepancy may be explained by ply interactions that are
simply not accounted for in the model. In the model assumptions section above, constant
§§ Poisson Effect – The phenomenon where a material experiences contraction in directions transverse to
the direction of applied tension and vice versa.
111
strain though the material’s thickness is explained; this assumption ties together elements
in the z-direction. This simplifies the model for computation time and development by
reducing the degrees of freedom within the model but with the now-obvious cost of
accuracy. Ply interactions in the experimental tests could include physical damage
occurring in mid-layer plies and inter-ply shear effects.
As mentioned above, this model was primarily used for supplementing and
interpreting experimental data. The primary process used was identifying damage modes
predicted by analytical failure criteria discussed above in the Multiaxial Testing section.
This provided analytical insight about the failure mode expected during the test and was
used to aid in the population of the progressive damage models displayed in APPENDIX
C.
112
CONCLUSIONS
First, to recap, the purpose of this research is to inform and validate progressive
damage models for their use with composite materials. Progressive damage models are
becoming more prevalent for use in designs and can provide crucial information about the
function and performance of a developing design. The ability to predict damage in any
material, especially composites has proven to be difficult and nearly impossible to do
accurately. There exist many analytical failure prediction models that have their own
strengths and weaknesses. To better understand the drawbacks of these analytical methods,
a large amount of test data is required. Furthermore, to inform and validate any model,
particularly progressive failure models, this requirement stands and experimentally
obtained data is crucial for their development and use.
Many of the analytical failure criteria combine mixed modes of loading in attempt
to predict damage. However, the validation of these mixed mode combinations remains
largely uncharacterized due to the inability for mixed mode testing. Typical mechanical
testing of materials includes tension, compression, shear, and many others but rarely
incorporates combinations such as biaxial tests. The Montana State University’s In-Plane
Loader poses a solution to this problem as it can apply combinations of displacements that
are atypical to standard mechanical tests. This study aimed to test composite materials in
many displacement combinations as an attempt to see how well multiaxial testing can
produce mixed-mode responses in the material.
In combination with mixed-mode testing, the use of digital image correlation for
full-field strain measurements was explored. The benefits of digital image correlation
113
rather than traditional strain measurement techniques, is that high resolution and low error
strain measurements can be collected for the full length of the test. Digital image
correlation is also helpful for the direct comparison, as mentioned several times, between
an experimental test and an analytical finite element model.
Experimental Conclusions
A lot of information can be gathered from the experimental data in this study. Even
though many of the results simply cannot be displayed in this document format, many of
them can provide the information desired to both qualitatively and quantitatively inform
and validate progressive damage models.
As described in the “Loading Paths” section, many different combinations of
displacements are applied to each set of coupons. This was done in attempt to fully
characterize the materials for many different loading cases. However, due to many factors,
a total of only one hundred and eleven (111) samples were used for this study to
characterize 6 independent laminates. This number of samples took a lot of time to test
and process, and for this reason, no more were attempted. Even with this number of
samples, the number of load paths was limited; for full characterization of a material, many
more tests should be performed. Furthermore, many tests of the same load path should be
run from a reliability standpoint and to increase the statistical significance of the data. For
further recommendations regarding experimental tests, refer to the recommendations
portion of this document.
Perhaps the two most useful results from this study are the qualitative video reports
created in ARAMIS and the failure surfaces for each laminate. First, the ARAMIS reports
114
offer the ability to compare, side by side, an experimental test and a finite element model
simulation of that test. This would primarily be used for the validation of a progressive
damage models and allow the user to determine if the model is accurately predicting
damage. Second, the failure surfaces produced may directly inform progressive damage
models as they provide experimental insight about when damage is predicted to occur
rather than various, and possibly inaccurate, analytical failure criteria.
The idea of producing experimental failure surfaces is simple. The strains
measured on a sample prior to damage are “acceptable” strains and do not cause failure.
However, when these strains exceed a limit, damage occurs. With digital image correlation,
the strains are collected in situ and the strains that cause failure can be isolated. These
strains are then displayed graphically in point clouds and a data density program determines
how reliable each individual point is. Regions that have many points are determined to be
“safe strains” where the material has repeatedly experienced these strains without damage
occurring. At some point, away from the cluster, the material is more likely to experience
damage; this is determined by the “empty” regions in the point cloud as the material has
never experienced strains of that magnitude without damaging.
The failure surfaces produced in this study “nest” several different calculated
surfaces which resemble levels of conservativeness. If the largest failure surface (red) is
chosen, it is obviously less conservative than the smallest failure surface (blue). A design
is less likely to incur damage when experiencing strains bounded by smaller surface than
the larger. Ultimately, these failure surfaces may be used to control damage models and
provide validation of existing and developing designs.
115
Since damage models typically require damage initiation criteria to function
accurately, these experimentally determined failure surfaces could be used to inform the
model as to when damage initiates. This likely would need to be done using a mathematical
representation of the failure surface rather than the amorphous experimental envelope.
This will be discussed further in the recommendations section to follow.
Furthermore, the failure surfaces produced may be useful for the validation/
qualification of existing designs. This may be done analytically or experimentally.
Analytically, the failure surface may be used to inform a model or be otherwise applied in
order to theoretical/ analytically determine strains of a design. Experimentally, measured
strains from a design incorporating the appropriate material may be compared to surface.
Strain data, whether measured or otherwise determined, that surpasses the boundaries of
the surface is predicted to have caused damage in the material. Using this information,
proper decisions can be made whether this means decommissioning an existing part,
increasing inspection intervals, repairing components, or redesigning the part.
Finally, the produced failure surfaces may be useful for the comparison of different
material systems. As displayed in Figure 36 above, different materials’ strain limits (and
stress limits) can be directly compared with each other. This may result in a useful tool for
the selection of materials in designs. If a material is selected, used in design, and
analytically/ experimentally determined to incur damage during service, a replacement for
the original material with better damage properties may be selected.
116
Finite Element Model Conclusions
Overall, the results produced from the model were determined to not suffice as a
method of informing and validating progressive damage models. Even though the model
was determined to not suffice in the primary purpose of this study, it did aid in the
interpretation of experimental results. There are several factors that contribute to this.
• Direct modeling of experimental tests.
o The method described above allows for the reproduction of experimental
data within a FEM with little error.
o Models are controlled by “exact” measurements determined by DIC.
• Digital image correlation has limits.
o Calculations of material responses under the outermost surface of the
coupon cannot be made with DIC. An FEM allows for inspection, on a
theoretical level, of material responses within each ply.
o Calculations cannot be defined on edges of a sample via DIC. FEMs can
produce results on these boundaries to supplement the experimental data.
o Although not used in this study, some FEMs (particularly damage models)
may provide insight about damage progressing through the material. DIC
typically produces large errors when examining a damaged sample.
• Applying analytical failure criteria.
o With the idealized reproduction of an experimental test, this data can be
analyzed further by applying analytical failure criteria. The model allows
for rapid adjustments within the model that apply many different failure
117
criteria with ease. This analysis would be lengthy and cumbersome without
features from the model.
o Analytical failure criteria, although perhaps inaccurate as described
previously, provide additional perspective on damage modes within a
sample.
It is going to be emphasized one last time that the model presented in this study is
not a progressive damage model. Even though this model incorporates a tool for applying
established failure criteria, the model does not attempt to reconcile damage prediction by
modeling the damage itself. The results of the model, discussed above, allow for several
conclusions to be made regarding this type of work and finite element models.
• Finite element models are useful for experimental data interpretation.
o Finite element models, no matter how idealized, may provide useful
analytical insight as to what occurs during an experimental test. Useful
interpretation tools could be:
▪ The application of analytical failure criteria to understand
damage.
▪ To simulate the response of materials where data can simply not
be collected during an experimental test e.g. edges of a sample
via digital image correlation.
▪ To determine how experimental results may stray from an
idealized analytical version.
• Finite element models are exactly that… models.
118
o Models require, on some level, assumptions to be made. As an example,
constitutive and damage responses characterizations never reduce into
the scalars that are used in models; variability always exists.
o The assumptions made in any model, no matter how complex, do not
and cannot fully describe what occurs in reality. These tools can
certainly be refined and honed to become more accurate but they should
be treated as the models that they are. Structures that utilize models in
their design process must be validated and substantiated by
experimental results.
119
FUTURE WORK
High Strain-Rate Multiaxial Testing
Very briefly, a new exploratory research project using the IPL to test materials under
high strain rates is being performed by Chris Stroili. This research is similar to the study
described in this document except that it focuses the use of multiaxial testing and digital
image correlation for the testing of isotropic materials. As the concept of multiaxial testing
is understood but rarely performed, this research focuses on discovering the merits of high
strain-rate multiaxial testing and its capabilities to produce data useful for modeling and
analysis.
Multiaxial Testing Recommendations
This portion will discuss the recommendations pertaining specifically to multiaxial
testing. A review of previous theses’ recommendations for the IPL clearly demonstrates
that the IPL is a “work in progress” and will likely remain as such. However, as problems
continue to be resolved, the requirements, justification, and utility for multiaxial testing
becomes more apparent.
Data Processing
There are many options that were not explored in this study to increase the
usefulness of this data. The topics below discuss a couple of the primary ideas for further
processing of the data.
• Comparison of experimental data to existing failure criteria
120
o The comparison of existing failure criteria and the experimentally
determined failure surfaces would not be to validate the experimental data.
This would only provide insight about how data should be interpreted and
used.
• Mathematical “fit” to experimental data
o An ellipsoidal (3-dimensional quadratic) fit program was written for this
study but ultimately not executed. The approximate run time for the script
was estimated in MATLAB to be over 50 hours.
▪ This MATLAB script employs a power least squares fit that weights
“outliers” less than highly dense data.
o Any mathematical (not necessarily quadratic) expression, may be useful for
the application of this data into a finite element software. Finite element
software can easily operate under mathematical parameters and the use of
the amorphous experimental envelope may prove impossible in a finite
element code.
Furthermore, in addition to ideas that may produce useful data, this short portion
will discuss recommendations about processing data in general. First, the data processed
in this study is dense and, to process, it requires a lot of computer resources. For this
reason, once data is processed, all variables should be saved so that their retrieval is easy.
If adjustments need to be made, existing variables can be used which will reduce processing
time. Second, as much processing as possible should be performed within ARAMIS
software before exporting. ARAMIS, especially newer versions, allow for calculations to
121
be made within the software. Any calculations that can be handled by ARAMIS should be
used and may even prove useful for creating additional qualitative reports like Figure 26
shown previously.
IPL Compliance
As discussed much in Smith’s thesis [13], the IPL compliance remains an issue
even up to the fifth generation of the IPL. Future work should determine a way to account
for the compliance in the IPL and/or increase its rigidity. In a perfect testing situation, the
machine used to provide loads to the coupon should be infinitely rigid or at least far more
rigid than the sample being tested. The IPL was designed sufficiently to handle loads from
the actuators through bearings, load cells, frame, fastener joints, and grips without
expecting permanent deformation of any sort to the IPL. However, even though the IPL is
designed sufficiently for strength, the components of the IPL act as springs and store elastic
energy that can then be transmitted to the coupon – particularly during damage. This poses
problems primarily from a control standpoint.
The IPL is a position controlled machine, as mentioned in its introduction. Since
the IPL acts as a spring and stores energy, this poses two primary problems for control of
the machine. First, an input for displacements is entered and the software tracks the
encoders on the actuators to ensure the prescribed displacement is achieved at the proper
rate, but the coupon doesn’t experience the prescribed displacement due to the deflection
of the IPL itself. This makes it difficult to achieve specific displacement states to the
coupon. Second, as the coupon experiences damage up to a critical value, the stored energy
in the IPL abruptly damages the coupon further. This is the primary issue caused by IPL
122
compliance as data over the abrupt damage period is nearly impossible to measure
accurately with the current equipment.
Automation of the IPL
Again, as mentioned in Smith’s thesis [13], tests require a substantial amount of
time investment. Previously, Smith mentions that tests take about a half hour to complete,
this remains to be the case due to the manual nature of the machine and data collection
necessary for each test. This does not include setup, intermittent calibrations, or teardown
of equipment. As mentioned earlier, it is necessary to have as many tests as possible to
produce statistically significant results. For this reason, it is recommended that future work
with the IPL be automated as much as possible – any reduction to test time would be
extremely beneficial. Furthermore, it is also recommended that future multiaxial testing
machine designs consider automation, specifically coupon loading and testing, as a design
priority.
IPL Software and Control
Although the software has been much improved due to the hard work of Dr. Michael
Edens, improvements could still be made. First, load control is currently being pursued as
an option of IPL control. Load control would consist of prescribing a load state and running
a PID (proportional, integral, and differential) control loop determining, in real-time,
necessary actuator displacements to achieve such a state. This would potentially mitigate
the rapid damage noted in the previous section and provide more versatility in testing.
Second, a more robust method of returning the IPL to a “home” position should be
considered. A faster “reset” between samples would cut down on test and setup time.
123
Figure 50. Third and Fourth generation IPL grip assembly as shown in Collett [11]. LVDTs
are used for the displacement control software
If the previously mentioned compliance issues remain uncorrected, a way of
potentially solving the resulting problems may be to instrument the IPL closer to the grips
for position control or feedback. Figure 50 shows a configuration of grip position data
acquisition that was used in the third and fourth generation IPL but was removed due to
signal noise issues in the LVDTs. The LVDTs were then, as mentioned before, appointed
for use as a means of quick calibration for the IPL position kinematics. Collett also
identified the position data acquisition at the grips to be inaccurate and proposed a change
in LVDT configuration like NRL’s in-plane loader. A new design should be implemented
for better control and validation of the IPL displacements whether a new, more accurate
124
LVDT setup such as that suggested by Collett or another form of accurate displacement
measurement technology (maybe even real-time DIC-IPL control).
Long-Term Suggestions
Whether employing a new multiaxial testing system entirely or considering major
retrofits to the existing IPL, here are several suggestions that are not discussed above. First,
consider hydraulic, servo-motor, or another form of high-speed actuation. Another
functionality of multiaxial testing could be fatigue testing or high strain rate testing for
multiaxial states which would require much faster and responsive actuation than the
existing stepper-motor actuators. An actuator retrofit to the existing IPL may be unfeasible
but for later versions of multiaxial testing machines, this should be a consideration.
Second, consider adding a dedicated third-party camera fixed to the IPL and
suitable for DIC measurements. This would reduce setup time and may mitigate
measurement issues when using DIC. As mentioned above, Parker attempted such a feat
using a MATLAB script [15], but it is easier to use the stereo camera setup for the existing
ARAMIS setup. Note: A single third-party camera with use on the ARAMIS cannot
capture three-dimensional deformations; and the use of a third-party camera will also
require nontrivial innovation to interface the IPL, camera, and ARAMIS systems.
Third, consider laying the IPL on its side similar to the first-generation IPL. This
would likely be better suited for an entirely new in-plane machine. Even with the new
‘out-of-plane constrainers’, maintaining proper cross-head alignment is difficult especially
during any compression test. One of the ideas discussed was mounting the cross-head to a
large air-bearing similar to what is used for three-axis machining with a large slewing
125
bearing to allow for rotation. This would give the proper degrees of freedom with only the
tolerances of the bearings, compliances, and joints to account for out-of-plane
displacements. The other (lower) part of the IPL would be fixed to the same planar surface
as the air-bearing.
Fourth, consider using other technologies in tandem with multiaxial testing.
Technologies that are readily available at MSU but have not been integrated into multiaxial
testing include but aren’t limited to: acoustic emission (AE), scanning electron microscopy
(SEM), CT scans, and ultrasound. These measurement technologies all provide different
means to quantify damage either during or post-test.
Digital Image Correlation (ARAMIS) Recommendations
The GOM ARAMIS digital image correlation system performs well when used
properly. However, since the system is about 8 years old, the software and hardware have
many expected performance issues. There are only several recommendations that should
be considered for any lengthy study involving the use of any digital image correlation
system.
Computer Resources
Computer resources including RAM, hard drive space, graphics cards, and CPU are
limited on this older system. These are listed with primary concerns first. RAM
occasionally poses a problem with processing large amounts of data and for the use of
“Fast-Measurement Mode”. Without an expansion of RAM, the system is limited on its
126
processing and measuring capabilities since much of the data is stored in RAM for
processing.
Second, hard drive space is an issue that should be handled if any large data sets
are desired. Stored tests require approximately 8 GB of storage space for 300 stages. For
the purpose of this study, most tests performed were stored as 150 stages to reduce this
storage cost but at a loss of data resolution. Finally, graphics cards, CPU, and perhaps
software limitations prevent the processing of large data sets. Each test may take
approximately 2 hrs. to process fully using this system. Much of this time spent is due to
a computer-only processing time that may be reduced with more efficient hardware.
Software Limitations
For this study, GOM ARAMIS v6.3 was used for the collecting and processing of
data. This version of the software limits users in some ways to automate tests and has
many limitations on processing abilities. It is highly recommended that, for future use with
this type of work, the latest software be used. The latest software boasts better automation
capabilities, exporting, and direct comparison within the software to finite element model
data. This can be used to correlate test data directly to a finite element model without the
use of a third-party software.
Camera Hardware
For this section, there are only a few recommendations. First, it is highly
recommended that calibration and setup remain the same as much as possible. Keeping
the cameras in a single location without moving them helps to reduce calibration issues.
Secondly, for the use of the existing cameras, additional well-filtered lighting should be
127
sought. The more light on the sample, the more closed the camera apertures can be which
increases depth of field and decreases shutter time. This will result in sharper images
especially for high frame-rates and high strain-rates.
Finally, new camera hardware should eventually be sought. This system has very
high-quality resolution; however, the latest models have more sensitive sensors that
perform better in low-light situations as well as record higher-resolution images. The
higher the image resolution, the better the strain field computation. The more sensitive the
sensor, the less light required and can therefore have increased shutter speed – ultimately,
this may allow for high strain-rate testing with the use of 3D digital image correlation.
128
REFERENCES CITED
129
[1] T. Palucka and B. Bensaude-Vincent, "Composites Overview," California
Technical Institute, 19 October 2002. [Online]. Available:
http://authors.library.caltech.edu/5456/1/hrst.mit.edu/hrs/materials/public/com
posites/Composites_Overview.htm. [Accessed 06 January 2017].
[2] National Materials Advisory Board, "Accelerating Utilization of New Materials,"
NMAB, Washington D.C., 1971.
[3] E. J. Barbero, Introduction to Composite Materials, 2nd ed., Boca Raton, FL:
Taylor and Francis Group, 2011.
[4] U.S. Department of Defense, MIL-HDBK-17-1F: Composite Materials
Handbook, Polymer Matrix Composites: Materials Usage, Design, and
Analysis, vol. 17, Department of Defense, 2002.
[5] Federal Aviation Administration, "de Havilland DH-106 Comet 1," U.S.
Department of Transportation, Washington, DC, 2016.
[6] T. L. Anderson, Fracture Mechanics - Fundamaentals and Applications 2nd
Edition, Boca Raton, Florida: CRC Press LLC, 1995.
[7] M. Hinton, A. S. Kaddour and P. D. Soden, Failure Criteria in Fibre Reinforced
Polymer Composites: The World-Wide Failure Exercise, Oxfod, UK: Elsevier,
2004.
[8] R. M. Christensen, "Failure Criteria," Stanford University, 9 June 2017. [Online].
Available: http://www.failurecriteria.com/theworldwidefail.html.
[9] P. W. Mast, G. E. Nash, J. G. Michopoulos, R. Thomas, R. Badaliance and I.
Wolock, "Characterization of strain-induced damaged in composites based on
the dissipated energy density," Theoretical and Applied Fracture Mechanics,
Vols. I, II, III, no. 22, pp. 71 - 125, 1995.
[10] E. Booth, K. Higgins and M. Schaff, "In-plane loader, a multi-axis composite
testing machine 'Senior Design Project'," Montana State University, Bozeman,
2001.
[11] A. B. Collett, ""A Validation Study of the Montana State University In-Plane
Loader" MSME thesis," Montana State University, Bozeman, 2006.
130
[12] W. Ritter, ""Application of Energy Methods to Modeling Failures in Composite
Materials and Structures" MSME thesis," Montana State University,
Bozeman, 2004.
[13] J. D. Smith, ""Internal Damage Characterization for Composite Materials Under
Biaxial Loading Configuration" MSME Thesis," Montana State University,
Bozeman, 2007.
[14] J. T. Schmitt, ""Damage Initiation and Post-Damge Response of Composite
Laminates by Multi-Axial Testing and Nonlinear Optimization" MSME
thesis," Montana State University, Bozeman, 2008.
[15] J. W. Parker, ""Development and Implementation of a Low cost Image Correlation
System to Obtain Full-Field In-Plane Displacement and Strain Data" MSME
thesis," Montana State University, Bozeman, 2009.
[16] F. Hild, A. Bouterf and S. Roux, "Damage Measurements via DIC From Physical
to Mechanical Damage," International Journal of Fracture, pp. 1-38, 2015.
[17] GOM mbH, ARAMIS v6.1 and higher User Manual, Braunschweig, Germany,
2007.
[18] Wichita State University, "Hexcel 8552 IM7 Unidirectional Prepreg 190 gsm &
35%RC Qualification Material Property Data Report," National Intitute for
Aviation Research, Wichita, KS, 2011.
[19] "McMASTER-CARR," McMASTER-CARR, [Online]. Available:
https://www.mcmaster.com/#.
[20] N. E. Dowling, Mechanical Behavior of Materials 'Engineering Methods for
Deformation, Fracture, and Fatigue', Fourth Edition ed., Upper Saddle River,
New Jersey: Pearson, 2013.
[21] GOM Aramis, a non-contact and material-independent measuring
software/hardware system based on digital image correlation. 49 531 390290,
www.gom.com.
[22] MATLAB, a mathematics and programming software available from Mathworks.
508 647 7000, www.mathworks.com.
[23] Instron, materials testing hardware and software available from INSTRON. 800
564 8378, www.instron.us/en-us/.
131
[24] NI LabVIEW, a commercial control software package available from National
Instruments. 877 388 1952, www.ni.com.
[25] ANSYS, a commercial finite element software package available from ANSYS Inc.
844 462 6797, www.ansys.com.
132
APPENDICES
133
APPENDIX A:
TEST MATRICES
134
Table 6. Test Matrix for MSU-11. [-45/90/45/0]s laminate with IM7/8552 material
system. Comp. abbreviation for compression
Test Name
Test
Speed
(in/min)
Control Displacements Damage
Initiation
Stage
Comments
dx (in) dy (in) d(deg)
11_003 0.1 -0.50 0.09 33.93 101 Preliminary test
11_004 0.1 -0.50 -0.03 44.30 72 Preliminary test
11_005 0.1 -0.50 0.09 33.93 90 Preliminary test
11_006 0.1 0.50 -0.19 -25.09 42 Preliminary test
11_007 0.1 -0.50 0.03 38.90 58 Preliminary test
11_008 0.1 -0.50 -0.03 44.30 62 Preliminary test
11_009 0.1 0.50 -0.19 -25.09 58 Preliminary test
11_010 0.05 -0.50 0.00 0.00 80
11_011 0.05 0.50 0.00 0.00 72
11_012 0.05 0.00 0.50 0.00 35
11_013 0.05 0.00 -0.50 0.00 Comp.- Not Used
11_014 0.2 0.00 0.00 -10.00 134
11_015 0.5 0.00 0.00 10.00 60
11_016 0.05 -0.50 0.50 0.00 44
11_017 0.05 0.50 0.50 0.00 62
11_018 0.2 -0.50 0.00 -10.00 104
11_019 0.2 0.50 0.00 -5.00 98
11_020 0.2 0.00 0.50 -10.00 69
11_021 0.2 0.00 0.50 10.00 68
11_022 0.4 -0.50 0.50 -10.00 56
11_023 0.4 0.50 0.50 -10.00 41
11_024 0.4 -0.50 0.50 10.00 32
11_025 0.4 0.50 0.50 10.00 63
135
Table 7. Test Matrix for MSU-13. [0/90/0/90]s laminate with IM7/8552 material system.
Comp. abbreviation for compression. NFF abbreviation for No Final Failure. GS
abbreviation for Grip Slippage
Test Name
Test
Speed
(in/min)
Control Displacements Damage
Initiation
Stage
Comments
dx (in) dy (in) d(deg)
13_008 0.05 0.50 0.00 0.00 105 Preliminary test
13_009 0.05 0.50 0.02 0.00 86 Preliminary test
13_010 0.1 -0.50 0.00 0.00 30
13_011 0.1 0.50 0.00 0.00 84
13_012 0.1 0.00 0.50 0.00 18 NFF - GS
13_013 0.1 0.00 -0.50 0.00 Comp.- Not Used
13_014 0.4 0.00 0.00 -10.00 54
13_015 0.8 0.00 0.00 10.00 41
13_016 0.1 -0.50 0.50 0.00 14
13_017 0.1 0.50 0.50 0.00 44
13_018 0.8 -0.50 0.00 -5.00 35
13_019 0.4 0.50 0.00 -5.00 45
13_020 0.5 0.00 0.50 -10.00 137 NFF - GS
13_021 0.8 0.00 0.50 10.00 23
13_022 0.6 -0.50 0.50 -10.00 22
13_023 0.6 0.50 0.50 -10.00 29
13_024 0.6 -0.50 0.50 10.00 20
13_025 0.6 0.50 0.50 10.00 43
136
Table 8. Test Matrix for MSU-14. [-45/45/-45/45]s laminate with IM7/8552 material
system. Comp. abbreviation for compression. NFF abbreviation for No Final Failure. GS
abbreviation for Grip Slippage. IPL Lim. Abbreviation for IPL Limit Reached
Test Name
Test
Speed
(in/min)
Control Displacements Damage
Initiation
Stage
Comments
dx (in) dy (in) d(deg)
14_010 0.1 -0.50 0.00 0.00 24
14_011 0.05 0.50 0.00 0.00 67
14_012 0.1 0.00 0.50 0.00 19
14_013 0.1 0.00 -0.50 0.00 Comp.- Not Used
14_014 0.5 0.00 0.00 -10.00 28 NFF - IPL Lim.
14_015 1 0.00 0.00 10.00 60 Buckle Failure
14_016 0.5 -0.50 0.50 0.00 12
14_017 0.1 0.50 0.50 0.00 44
14_018 0.1 -0.50 0.00 -10.00 42 NFF - IPL Lim.
14_019 0.5 0.50 0.00 -5.00 46 NFF - GS
14_020 0.5 0.00 0.50 -10.00 28
14_021 0.5 0.00 0.50 10.00 41
14_022 0.4 -0.50 0.50 -10.00 51
14_023 0.4 0.50 0.50 -10.00 58
14_024 0.4 -0.50 0.50 10.00 30
14_025 0.6 0.50 0.50 10.00 27
137
Table 9. Test Matrix for MSU-1. [-45/90/45/0]s laminate with Toray material system.
Comp. abbreviation for compression. NFF abbreviation for No Final Failure. GS
abbreviation for Grip Slippage
Test Name
Test
Speed
(in/min)
Control Displacements Damage
Initiation
Stage
Comments
dx (in) dy (in) d(deg)
1_005 0.05 0.00 0.50 0.00 53 Preliminary test
1_006 0.03 0.00 0.50 0.00 46 Preliminary test
1_009 1 0.13 0.13 5.00 31 Preliminary test
1_010 0.05 -0.50 0.00 0.00 43
1_011 0.05 0.50 0.00 0.00 121 NFF - GS
1_012 0.05 0.00 0.50 0.00 41 Noisy Data
1_013 0.05 0.00 -0.50 0.00 Comp.- Not Used
1_014 0.2 0.00 0.00 -10.00 175
1_015 0.5 0.00 0.00 10.00 26
1_016 0.05 -0.50 0.50 0.00 36
1_017 0.1 0.50 0.50 0.00 56
1_018 0.2 -0.50 0.00 -5.00 89
1_019 0.2 0.50 0.00 -5.00 31
1_020 0.5 0.00 0.50 -10.00 226 NFF - GS
1_021 0.5 0.00 0.50 10.00 34
1_022 0.4 -0.50 0.50 -10.00 54
1_023 0.4 0.50 0.50 -10.00 63
1_024 0.4 -0.50 0.50 10.00 23
1_025 0.4 0.50 0.50 10.00 46
1_026 0.1 0.50 0.50 0.00 58
138
Table 10. Test Matrix for MSU-3. [0/90/0/90]s laminate with Toray material system.
Comp. abbreviation for compression. NFF abbreviation for No Final Failure. GS
abbreviation for Grip Slippage. IPL Lim. Abbreviation for IPL Limit Reached
Test Name
Test
Speed
(in/min)
Control Displacements Damage
Initiation
Stage
Comments
dx (in) dy (in) d(deg)
3_007 0.08 0.00 0.50 0.00 32 Preliminary test
3_008 0.04 0.00 0.50 0.00 34 Preliminary test
3_009 1 -0.25 0.25 10.00 26 Preliminary test
3_010 0.1 -0.50 0.00 0.00 43
3_011 0.1 0.50 0.00 0.00 61
3_012 0.1 0.00 0.50 0.00 28 NFF - GS
3_013 0.1 0.00 -0.50 0.00 Comp.- Not Used
3_014 0.8 0.00 0.00 -10.00 65
3_015 0.8 0.00 0.00 10.00 97
3_016 0.1 -0.50 0.50 0.00 20
3_017 0.1 0.50 0.50 0.00 16 NFF - GS
3_018 0.8 -0.50 0.00 -5.00 7 Fast Test
3_019 0.4 0.50 0.00 -5.00 65
3_020 0.8 0.00 0.50 -10.00 44 NFF - IPL Lim.
3_021 0.8 0.00 0.50 10.00 24
3_022 0.6 -0.50 0.50 -10.00 46 NFF - IPL Lim.
3_023 0.6 0.50 0.50 -10.00 33 NFF - IPL Lim.
3_024 0.6 -0.50 0.50 10.00 24
3_025 0.6 0.50 0.50 10.00 45
139
Table 11. Test Matrix for MSU-3. [-45/45/-45/45]s laminate with Toray material system.
Comp. abbreviation for compression. NFF abbreviation for No Final Failure. GS
abbreviation for Grip Slippage. IPL Lim. Abbreviation for IPL Limit Reached
Test
Name
Test
Speed
(in/min)
Control Displacements Damage
Initiation
Stage
Comments
dx (in) dy (in) d(deg)
4_005 0.06 0.00 0.50 0.00 48 Preliminary test
4_006 0.06 0.00 0.50 0.00 16 Preliminary test
4_007 0.06 0.00 0.50 0.00 43 Preliminary test
4_008 0.06 0.00 0.50 0.00 58 Preliminary test
4_009 1 -0.25 0.25 10.00 64 Preliminary test
4_010 0.05 -0.50 0.00 0.00 36
4_011 0.05 0.50 0.00 0.00 56
4_012 0.2 0.00 0.50 0.00 9
4_013 0.2 0.00 -0.50 0.00 Comp.- Not Used
4_014 1 0.00 0.00 -10.00 3 NFF - IPL Lim.
4_015 1 0.00 0.00 10.00 87 NFF - IPL Lim.
4_016 0.2 -0.50 0.50 0.00 14
4_017 0.1 0.50 0.50 0.00 39
4_018 0.5 -0.50 0.00 -5.00 33
4_019 0.5 0.50 0.00 -5.00 71 NFF - GS
4_020 0.5 0.00 0.50 -10.00 42
4_021 0.5 0.00 0.50 10.00 47
4_022 0.4 -0.50 0.50 -10.00 31 NFF - IPL Lim.
4_023 0.5 0.50 0.50 -10.00 30
4_024 0.5 -0.50 0.50 10.00 40
4_025 0.5 0.50 0.50 10.00 35
140
APPENDIX B:
LAMINATE-LEVEL FAILURE
141
Figure 51. Nested isosurfaces for MSU - 13 displayed on each primary plane. Original
nested isosurfaces (top-left). Primary plane eps xy = 0 (top-right). Primary plane eps y =
0 (bottom-left). Primary plane eps x = 0 (bottom-right)
142
Figure 52. Nested isosurfaces for MSU - 14 displayed on each primary plane. Original
nested isosurfaces (top-left). Primary plane eps xy = 0 (top-right). Primary plane eps y =
0 (bottom-left). Primary plane eps x = 0 (bottom-right)
143
Figure 53. Nested isosurfaces for MSU - 3 displayed on each primary plane. Original
nested isosurfaces (top-left). Primary plane eps xy = 0 (top-right). Primary plane eps y =
0 (bottom-left). Primary plane eps x = 0 (bottom-right)
144
Figure 54. Nested isosurfaces for MSU - 4 displayed on each primary plane. Original
nested isosurfaces (top-left). Primary plane eps xy = 0 (top-right). Primary plane eps y =
0 (bottom-left). Primary plane eps x = 0 (bottom-right)
145
APPENDIX C:
PROGRESSIVE DAMAGE TABLE
146
Each damage mode in the following tables is determined solely based on ARAMIS
calculations such as major strain (maximum principal strain) direction and the observation
of failed samples. The data was then interpreted analytically using the finite element
model. All damage modes included are only damage modes that could be directly observed
from the data. For this reason, some listed failure modes were broad and not specifically
defined. Abbreviations for the interpretation of the tables are as follows:
• MC – Matrix Cracking
o Cracks begin in the matrix, propagates, and leaves fibers intact.
Typically observed as the first damage. High strains and maximum
principal strain directions provide information regarding this type of
damage. Further details into this mode of failure (such as shear or
tension) are not easily determined and therefore not included in the
resultant tables.
• DL – Delamination
o Often interpreted from out-of-plane displacements calculated by the
DIC. This failure typically causes final failure for tests with large
rotation components as fiber failure is observed for these
displacements less often.
• FR – Fiber Rupture
o Fibers incur damage directly causing fiber breakage. Typically, the
“last straw” as stored energy suddenly released into the sample
causes catastrophic failure.
• Buckle – Buckle Failure
o Due to compression, typically causes large out-of-plane
displacements, delaminations, and even kink-band formation.
• OA – Off Angle Plies
o All non-zero and non-90 degree plies.
• LNT – Localized at Notch Tip
o Damage location is confined primarily to the notch tip. If not
specified, damage occurs multiple places.
• TG|BG – Top Grip or Bottom Grip
o Damage is occurring locally at the top grip or bottom grip.
147
Table 12. Progressive damage table for MSU–11 tests. IM7/8552 Material: [-45/90/45/0]s
stage ply mode stage ply mode stage ply mode
11_003 101 -45 MC - LNT 128 All DL 143 0 FR - LNT
11_004 72 90 MC - LNT 84 45 MC - LNT 125 0 FR - LNT
11_005 90 -45 MC - LNT 95 90 MC 107 All DL
11_006 42 -45 MC - LNT 61 90 MC 153 0 FR - LNT
11_007 58 0 MC 85 -45 MC - LNT 112 All DL
11_008 62 90 MC - LNT 86 OA MC - LNT 105 All DL
11_009 58 45 MC 148 -45 DL 283 All DL
11_010 80 90 MC - LNT 124 -45 DL 141 All DL
11_011 72 45 MC - LNT 97 OA MC - LNT 116 All DL
11_012 35 90 MC - LNT 77 OA MC 144 0 FR
11_014 134 90 MC 162 OA MC 301 All DL - LNT
11_015 60 90 MC 121 -45 MC - TG 204 All Buckle
11_016 44 90 MC - LNT 125 OA MC - LNT 175 0 FR - LNT
11_017 62 90|-45 MC - LNT 124 OA MC - LNT 179 0 FR - LNT
11_018 104 90 MC 167 90 MC - LNT 220 All DL - LNT
11_019 98 45 MC 180 OA MC 208 All DL
11_020 69 90 MC 163 OA MC 301 0 FR
11_021 68 90 MC - LNT 145 OA MC - LNT 212 0 FR - LNT
11_022 56 OA MC - LNT 100 90 MC - LNT 130 0 FR
11_023 41 90 MC 75 -45 MC - LNT 126 0 FR
11_024 32 90 MC - LNT 56 OA MC - LNT 154 0 FR - LNT
11_025 63 45 MC - LNT 69 -45 FR 74 All DL
Damage initiation Intermittent Stage Final FailureTest
Name
148
Table 13. Progressive damage table for MSU–13 tests. IM7/8552 Material: [0/90/0/90]s
stage ply mode stage ply mode stage ply mode
13_008 105 90 MC - LNT 180 0 MC - LNT 218 0 FR
13_009 86 0 MC - LNT 178 0 FR 234 0 FR - BG
13_010 30 0 MC - LNT 113 0 FR 134 0 FR - TG
13_011 84 0 MC - LNT 116 90 MC 123 0 FR
13_012 18 90 MC - LNT 38 0 MC - LNT
13_014 54 90 MC - LNT 137 0 MC - LNT 225 0 FR
13_015 41 90 MC 50 90 MC - LNT 63 0 FR - LNT
13_016 14 90 MC - LNT 72 90 MC 129 0 FR - LNT
13_017 44 90 MC 81 All DL 164 0 FR
13_018 35 All MC 55 0 FR 71 All DL
13_019 45 90 MC - BG 69 0 MC - LNT 175 0 FR - BG
13_020 137 90 MC - LNT 159 0 MC - LNT
13_021 23 90 MC 38 0 MC 73 0 FR - LNT
13_022 22 90 MC - LNT 88 0 MC - LNT 130 0 FR - LNT
13_023 29 0 MC 41 90 MC - LNT
13_024 20 90 MC - LNT 31 0 MC - LNT 89 0 FR - LNT
13_025 43 0 MC 77 90 MC - LNT 92 0 FR
Test
Name
Damage initiation Intermittent Stage Final Failure
149
Table 14. Progressive damage table for MSU–14 tests. IM7/8552 Material: [-45/45/-
45/45]s
stage ply mode stage ply mode stage ply mode
14_010 24 45 MC - LNT 49 -45 MC 81 All DL
14_011 67 -45 MC - LNT 108 -45 MC 150 45 FR
14_012 19 All MC - LNT 63 -45 FR - LNT 98 All DL
14_014 28 -45 MC - LNT 126 -45 MC - BG
14_015 60 -45 MC - LNT 91 45 MC 126 All Buckle
14_016 12 -45 MC 17 All MC 22 All DL
14_017 44 -45 MC - LNT 57 -45 MC 90 All DL
14_018 42 -45 MC - LNT 65 45 MC - LNT
14_019 46 All MC - LNT 100 All MC
14_020 28 All MC 100 -45 MC 152 All DL
14_021 41 -45 MC - LNT 81 All MC 129 All DL
14_022 51 -45 MC - LNT 154 45 MC 194 All DL
14_023 58 -45 MC 67 45 MC - LNT 132 All DL
14_024 30 All MC 98 All DL 167 -45 FR
14_025 27 -45 MC 62 OA MC - LNT 76 All DL
Test
Name
Damage initiation Intermittent Stage Final Failure
150
Table 15. Progressive damage table for MSU–1 tests. Toray Material: [-45/90/45/0]s
stage ply mode stage ply mode stage ply mode
1_005 53 90 MC 65 OA MC 112 All DL
1_006 46 90 MC - LNT 67 OA MC - LNT 92 0 FR - LNT
1_009 31 90 MC - LNT 60 OA MC - LNT 90 All DL
1_010 43 -45 MC - LNT 106 45|90 MC 113 All DL
1_011 121 90 MC - LNT 182 OA MC - LNT
1_012 41 -45 MC 83 All DL - LNT 178 0 FR - LNT
1_014 175 90 MC - LNT 232 OA MC - LNT 298 All DL
1_015 26 90 MC - LNT 46 OA MC - TG
1_016 36 90 MC - LNT 117 OA MC - LNT
1_017 56 90 MC 70 OA MC 192 0 FR
1_018 89 90 MC - LNT 111 OA MC - LNT 124 All DL - LNT
1_019 31 -45 MC 40 All DL 290 45 FR
1_020 226 90 MC - LNT 281 OA MC - LNT
1_021 34 90 MC - LNT 62 45 MC - LNT 79 0 FR - LNT
1_022 54 90 MC 80 OA MC 195 All DL
1_023 63 90 MC - LNT 97 OA MC 166 0 FR - LNT
1_024 23 90 MC - LNT 55 OA MC 103 All DL
1_025 46 90 MC - LNT 62 OA MC 101 0|45 FR
1_026 58 90 MC - LNT 72 OA MC 94 0 FR - LNT
Test
Name
Damage initiation Intermittent Stage Final Failure
151
Table 16. Progressive damage table for MSU–3 tests. Toray Material: [0/90/0/90]s
stage ply mode stage ply mode stage ply mode
3_007 32 90 MC 51 All DL 82 0 FR - LNT
3_008 34 90 MC - LNT 62 0 MC 75 0 FR - LNT
3_009 26 90 MC - LNT 45 0 FR - LNT 68 0 FR
3_010 43 0 MC - LNT 105 90 MC - LNT 147 0 FR
3_011 61 0 MC - LNT 141 90 MC - BG 178 All DL - LNT
3_012 28 90 MC - LNT 170 90 MC
3_014 65 90 MC - LNT 100 0 FR - LNT 151 0 FR
3_015 97 90 MC 110 All DL 122 0 FR
3_016 20 90 MC - LNT 80 All MC 139 0 FR - LNT
3_017 16 90 MC 160 0 MC - LNT
3_018 7 All MC 21 0 FR - LNT 24 0 FR - TG
3_019 65 All MC 152 All DL 195 0 FR
3_020 44 90 MC - LNT 140 0 MC - LNT
3_021 24 90 MC - LNT 62 0 MC - LNT 94 0 FR - LNT
3_022 46 0 MC - LNT 120 90 MC
3_023 33 0 MC - LNT 127 90 MC
3_024 24 90 MC - LNT 78 0 MC - LNT 128 0 FR - LNT
3_025 45 All MC 85 All DL 107 0 FR
Test
Name
Damage initiation Intermittent Stage Final Failure
152
Table 17. Progressive damage table for MSU–4 tests. Toray Material: [-45/45/-45/45]s
stage ply mode stage ply mode stage ply mode
4_005 48 -45 MC - LNT 151 45 MC 209 All DL
4_006 16 45 MC - LNT 120 -45 MC - LNT
4_007 43 -45 MC - LNT 150 45 MC 200 All DL
4_008 58 45 MC - LNT 148 -45 MC 205 All DL
4_009 64 -45 MC - LNT 135 All DL 149 -45 Buckle
4_010 36 45 MC - LNT 87 All DL 109 -45 FR
4_011 56 -45 MC - BG 108 45 MC 121 All DL
4_012 9 -45 MC - LNT 57 45 MC - LNT 65 All DL
4_014 3 45 MC - LNT 114 -45 MC - LNT
4_015 87 All MC - LNT 206 All MC
4_016 14 All MC 43 All DL 59 All DL
4_017 39 45 MC 88 -45 MC - LNT 136 All DL
4_018 33 45 MC - LNT 56 45 MC 68 -45 FR
4_019 71 45 MC - LNT 131 All MC
4_020 42 All MC - LNT 150 All MC 206 All DL
4_021 47 -45 MC - LNT 79 45 MC - LNT 171 All DL
4_022 31 All MC - LNT 184 All MC
4_023 30 -45 MC - LNT 121 All MC 152 All DL
4_024 40 All MC - LNT 125 All MC 173 All DL
4_025 35 -45 MC - LNT 71 45 MC 110 All DL
Test
Name
Damage initiation Intermittent Stage Final Failure